MulitUI
Collection
MultiUI: 7M multimodal UI instructions
•
5 items
•
Updated
•
7
id
stringlengths 10
13
| image
dict | conversations
listlengths 2
2
|
|---|---|---|
mind2web_0 | {
"bytes": "",
"path": "./images/ff6ff6e5-e73c-4ba2-809e-59a1e269d68f_c7d6e34d-c623-4e8a-93f4-c19e1269d82d.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find Hair Loss Evaluation and Treatment in 10018 and show directions.\nPrevious actions:\n[button] Health -> CLICK\n[button] load Health Services Menu -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.197, 0.492, 0.504, 0.522] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_1 | {
"bytes": "",
"path": "./images/34e13beb-0235-41d4-b108-137cc7480904_01b5c5b4-5304-4c38-9e8c-36cf97ccfa74.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book the cheapest parking spot for a standard car near 66 Perry St., New York, from April 20, 5:30 pm to April 21, 5:30 am.\nPrevious actions:\n[link] New York -> CLICK\n[link] New York -> CLICK\n[span] Enter an Address or Venue to find parking -> TYPE: 66 perry st\n[span] 66 -> CLICK\n[span] Apr 18 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.665, 0.312, 0.694, 0.346] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_2 | {
"bytes": "",
"path": "./images/8f4a020d-ac81-4d20-ae3a-02808faaba4e_77eb7f8e-0e43-4535-b87b-9704908c779f.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Buy a birthday party package for a group of 10 at Arizona park with my address details. Group name is Neo, group leader's name is Thomas Anderson, birthday is May 5th, 1995. The phone number is2526545258. The email address is thomas.neo@gmail.com and address is po box 2846, Texas city in the state of Texas.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.515, 0.693, 0.783, 0.755] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_3 | {
"bytes": "",
"path": "./images/521d9006-4560-49af-b232-c713d87dd2e2_84fd0d10-e222-4c1b-a852-e49ac16d3462.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Buy a group ticket for 20 people in Six Flags Magic Mountain in Los Angeles on Apr 30, add 5 mega meal deals with it. Name the group Crew and check out as James Johnson, james.john@gmail.com. It is a family trip organized by Johnson.\nPrevious actions:\n[button] Browse the Parks Below -> CLICK\n[span] Six Flags Magic Mountain -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.783, 0.445, 0.843, 0.485] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_4 | {
"bytes": "",
"path": "./images/b5c98548-4a3f-4e7c-8287-c36963930348_e3e58722-6d08-46af-bb39-109c07dc6874.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get a rental car to pick up at Paris Charles de Gaulle Airport CDG on april 13 at 10 am and drop off on april 18 at the same time\nPrevious actions:\n[div] Car Hire -> CLICK\n[textbox] I need a car in -> TYPE: Paris Charles de Gaulle Airport (CDG)\n[button] Paris Charles de Gaulle Airport (CDG) Paris, Franc... -> CLICK\n[generic] 13 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.324, 0.255, 0.346, 0.269] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_5 | {
"bytes": "",
"path": "./images/0b2c1886-0345-4ca7-a555-8b1cd1b1c40a_670d6d1f-a265-4b87-bb85-bcbe74cf3740.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse certified fresh sci-fi films streaming at home.\nPrevious actions:\n[link] MOVIES -> HOVER"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.253, 0.339, 0.424, 0.351] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_6 | {
"bytes": "",
"path": "./images/c3e841d5-4624-44cb-a8b5-897b9aa3ef9b_a16a9f27-7699-4f5f-a78f-5b0d6429569c.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Buy a single day pass to Six Flags, Magic Mountain.\nPrevious actions:\n[button] Browse the Parks Below -> CLICK\n[span] Six Flags Magic Mountain -> CLICK\n[button] Go! -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.312, 0.068, 0.489, 0.08] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_7 | {
"bytes": "",
"path": "./images/bb3e2b61-c14c-4a15-8de4-7c12256d52c7_b17d8b15-9af5-4661-9d7b-74851a227b83.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for concerts taking place in Los Angeles, CA in the next month.\nPrevious actions:\n[button] Change Location -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.097, 0.619, 0.231, 0.631] in the image\nAnd my action is Action: TYPE\nValue: Los Angeles"
}
] |
mind2web_8 | {
"bytes": "",
"path": "./images/844f8d77-e12e-4d68-8c4c-9146bd5b1530_f7ca543f-c1ac-4a4f-9bf1-f9980a41a07a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Open gift list for beauty products.\nPrevious actions:\n[path] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.0, 0.045, 0.249, 0.06] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_9 | {
"bytes": "",
"path": "./images/4272e233-3440-4572-bd86-b3a2b22a4061_515c6c84-9b18-49a5-a48c-2bdb562a48c3.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the most popular tours and attractions in Kyoto, Japan from May 25 with free cancellation option, check availability for 2 adults and 1 infant, and book.\nPrevious actions:\n[svg] -> CLICK\n[checkbox] 25 May 2023 -> CLICK\n[span] -> CLICK\n[span] -> CLICK\n[span] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.563, 0.134, 0.74, 0.15] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_10 | {
"bytes": "",
"path": "./images/2f660153-d470-4744-8b75-f8dc64c7f94e_1b4aa789-a458-4655-9eb9-f9e72cb900fd.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Explore Games & Puzzles in the Toys category that are suitable for ages 14+\nPrevious actions:\n[span] Toys -> CLICK\n[span] Games & Puzzles -> CLICK\n[button] Explore All -> CLICK\n[button] Age -> CLICK\n[div] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.969, 0.272, 0.981, 0.281] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_11 | {
"bytes": "",
"path": "./images/0c7a69f9-989d-4899-a3fe-94c952859007_83cd5543-cd4f-4bfe-9d7e-4ce6bf0dce32.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find flights from Chicago to London on 20 April and return on 23 April.\nPrevious actions:\n[a] CHI - Chicago, IL -> CLICK\n[textbox] To -> TYPE: London\n[a] LON - London, United Kingdom -> CLICK\n[button] View Calendar: To navigate the calendar, use the... -> CLICK\n[link] 20 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.599, 0.396, 0.614, 0.407] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_12 | {
"bytes": "",
"path": "./images/3e50b35f-7af9-4835-a741-158c2d941722_97174dd3-20af-42da-b81d-bed5eadb1b29.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the cheapest room for Athens, Attica, Greece 1 Apr 2023 - 3 Apr 2023, 2 nights, 1 Room, 1 Adult\nPrevious actions:\n[div] Choose date -> CLICK\n[generic] 3 -> CLICK\n[button] Search -> CLICK\n[button] Yes, I agree -> CLICK\n[span] Lowest price -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.742, 0.162, 0.855, 0.179] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_13 | {
"bytes": "",
"path": "./images/a6372f23-f462-4706-8455-5b350c46d83c_dab1ffd9-2df1-450b-8d68-120b605f8d45.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book a winery tour in Napa Valley in a winery which serves Mediterranean cuisine with wine testing for 4 guests on April 15, 10 am in a outdoor setup.\nPrevious actions:\n[svg] -> CLICK\n[svg] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.281, 0.108, 0.719, 0.149] in the image\nAnd my action is Action: TYPE\nValue: NAPA VALLEY"
}
] |
mind2web_14 | {
"bytes": "",
"path": "./images/4faf56f2-217b-4d2d-b510-a24e30e6b20e_7c992448-8fd4-4aff-96a6-34b790363b71.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a list of all movies that are being released in May 2023\nPrevious actions:\n[button] Reviews -> CLICK\n[button] HOME -> CLICK\n[button] Discover -> CLICK\n[link] Upcoming Movies -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.434, 0.118, 0.535, 0.153] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_15 | {
"bytes": "",
"path": "./images/dd13bc51-f582-4004-8641-eb4e62cabfc7_024c93e9-5579-44a1-bf6a-c773f34b8d34.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get prequalified with a good credit score for a $30000 car with a $6000 down payment for 48 months.\nPrevious actions:\n[menuitem] Finance -> HOVER\n[menuitem] Prequalify for Financing -> CLICK\n[input] -> TYPE: 30000\n[input] -> TYPE: 6000\n[span] Good -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.151, 0.583, 0.259, 0.603] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_16 | {
"bytes": "",
"path": "./images/c5070129-37b1-4f87-930f-c45049a70e97_2e80fca7-6d2c-46fa-a7f3-a9569f0dc82d.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Use the FareFinder to find flights for $150 or less from Dublin to anywhere.\nPrevious actions:\n[button] Plan -> CLICK\n[link] Fare Finder -> CLICK\n[textbox] From: -> CLICK\n[div] Dublin -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.356, 0.224, 0.553, 0.25] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_17 | {
"bytes": "",
"path": "./images/bd2b5866-dced-4ec1-8b3e-8a8a24e71a96_d3e3e41f-c601-4798-b602-6990777eba4a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Check all English Reading and Writing skills Educational Books for Ages 6-8 currently in stock that are available in hardback under 20$.\nPrevious actions:\n[link] Shop by category∨ -> CLICK\n[link] Shop by category∨ -> CLICK\n[link] Ages 6-8 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.038, 0.205, 0.196, 0.222] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_18 | {
"bytes": "",
"path": "./images/4af615be-4472-4357-8737-5286b06b385d_7fd3adeb-d440-4a31-a75e-3d2c009e00b2.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse listings for a used 2006 Honda Civic.\nPrevious actions:\n[link] Price New/Used -> CLICK\n[combobox] Year -> SELECT: 2006\n[combobox] Make -> SELECT: Honda"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.278, 0.306, 0.508, 0.34] in the image\nAnd my action is Action: SELECT\nValue: Civic"
}
] |
mind2web_19 | {
"bytes": "",
"path": "./images/46a3683f-fbe0-40d0-8729-6c7964d994e6_7425e091-4439-49a8-a6ff-c355ec0c4f34.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a men's UA outlet T-shirt of XL size and add to cart.\nPrevious actions:\n[link] Short Sleeves -> CLICK\n[div] Size -> CLICK\n[link] XL -> CLICK\n[img] Men's UA Tech™ 2.0 Short Sleeve -> CLICK\n[button] XL -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.664, 0.19, 0.952, 0.209] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_20 | {
"bytes": "",
"path": "./images/b1055658-7993-448c-9af6-a722cf69ff97_2255c0f5-0829-4874-89d2-1dafcd92da6a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Make a reservation for the next restaurant-related event, only 1 ticket is needed.\nPrevious actions:\n[link] See more venues in the Special Events & Experi... -> CLICK\n[link] HAPPY HOUR BURGER, Aria, Thursday, April 13, 2023 ... -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.055, 0.256, 0.159, 0.278] in the image\nAnd my action is Action: SELECT\nValue: 1 Ticket"
}
] |
mind2web_21 | {
"bytes": "",
"path": "./images/38fe67f7-14af-4259-8309-aa350abdc395_f2fd4cab-af21-4062-bc3e-9899832b6611.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for the cheapest two rooms near Kashi Vishwanath Temple in India for three adults and a 7-year-old kid from June 6 to 10 in any 3-star and up air-conditioned family hotel with a review score of at least 8, free internet.\nPrevious actions:\n[path] -> CLICK\n[div] 8+ -> CLICK\n[div] 3 -> CLICK\n[checkbox] Free internet -> CLICK\n[button] Style -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.071, 0.825, 0.081, 0.832] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_22 | {
"bytes": "",
"path": "./images/59bdd427-c34b-4b93-a8b8-0c3a49acdcfa_e9d3601e-1394-445d-b999-b957959694a5.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book a round-trip flight from Santa Fe, New Mexico to Phoenix, Arizona for 1 adult on August 13th and returning on August 28th.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.075, 0.217, 0.259, 0.246] in the image\nAnd my action is Action: TYPE\nValue: Santa Fe"
}
] |
mind2web_23 | {
"bytes": "",
"path": "./images/913b902b-37b7-4230-a0f2-a19931b06951_04f0418a-4171-42ec-9588-fbf470f54df0.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the fastest single stop flight fron JFK airport in New York to Male in Maldives on March 13\nPrevious actions:\n[span] Round-trip -> CLICK\n[tab] One-way -> CLICK\n[textbox] Flight origin input -> TYPE: new york\n[div] John F Kennedy Intl -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.514, 0.102, 0.702, 0.126] in the image\nAnd my action is Action: TYPE\nValue: male"
}
] |
mind2web_24 | {
"bytes": "",
"path": "./images/a531a379-7770-4b72-a8cc-ee19419536cb_76b93c24-ab52-4865-8491-2c3423d615af.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: search for art galleries in midtown\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.451, 0.07, 0.519, 0.081] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_25 | {
"bytes": "",
"path": "./images/b467797c-5fab-4ee6-9b1f-ac6dc50c05cf_3cb50019-9056-4144-9449-be80b231f3cd.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the lowest-priced single pack of Xerox genuine magenta toner sold by Newegg with free shipping.\nPrevious actions:\n[link] Printer Ink & Toner -> CLICK\n[link] Xerox Toner -> CLICK\n[span] Newegg -> CLICK\n[button] APPLY -> CLICK\n[span] Magenta -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.148, 0.114, 0.192, 0.127] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_26 | {
"bytes": "",
"path": "./images/4bce534f-3057-40ca-aecd-ffb07b3c9fcb_b7a36124-3c4f-4a40-b927-9e0c1f548427.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search the actor who won the Oscars for best actor in a suuporing role ine 1990, and add his upcoming movies to my watchlist.\nPrevious actions:\n[button] Open Navigation Drawer -> CLICK\n[span] Oscars -> CLICK\n[link] 1990 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.196, 0.181, 0.289, 0.191] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_27 | {
"bytes": "",
"path": "./images/cd4df5fa-2c90-4a76-b4b0-0ca7bcdac2ee_6c1f5f61-3aa6-4eba-bd8b-ef20145ac9b9.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Reserve a train ticket with seat from Berlin to Zurich on March 31 after 8 am in aisle quite zone in side by side seat formation.\nPrevious actions:\n[span] Zürich -> CLICK\n[textbox] Date use format: 25-Mar-23 -> CLICK\n[link] 31 -> CLICK\n[listbox] hour -> SELECT: 08\n[listbox] minutes -> SELECT: 00"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.074, 0.342, 0.327, 0.371] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_28 | {
"bytes": "",
"path": "./images/56cfe20a-f008-48d3-b683-002ce7790616_19e36673-154e-407f-9425-7d8c2dfdd30c.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find cheapest luxury car rental in New York for two with pickup from JFK airport at 9am, April 3 and drop off at JFK on 6pm, April 6\nPrevious actions:\n[tab] Cars -> CLICK\n[button] Pick-up -> CLICK\n[textbox] Pick-up -> TYPE: jfk\n[button] New York (JFK - John F. Kennedy Intl.) New York, U... -> CLICK\n[button] Same as pick-up -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.505, 0.181, 0.931, 0.227] in the image\nAnd my action is Action: TYPE\nValue: jfk"
}
] |
mind2web_29 | {
"bytes": "",
"path": "./images/0f9dd411-f3ab-494e-8a3a-5dc9a908f4ee_cad3fd8c-22ca-4bef-806e-3ffa533fa0b0.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Locate the cvs pharmacy nearest to zip code 45201 that provides hair loss evaluation and treatment.\nPrevious actions:\n[link] Schedule a women's health exam -> CLICK\n[link] Explore all MinuteClinic health services -> CLICK\n[p] Skin, Hair & Nails -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.335, 0.286, 0.464, 0.295] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_30 | {
"bytes": "",
"path": "./images/a2b1d94b-5309-4c9b-bcdd-8048172af7b5_39e4017e-d59e-4582-be70-07a8b8cfd2fa.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find an foldable over-ear headphone under 35 USD in certified refurbished condition with lowest price + shipping.\nPrevious actions:\n[link] Electronics -> HOVER\n[link] Headphones -> CLICK\n[link] Ear-Cup (Over the Ear) -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.848, 0.141, 0.908, 0.154] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_31 | {
"bytes": "",
"path": "./images/63529dc2-7b13-493e-9a0c-b1ce50256a16_9e6abcf6-1bbe-42c9-bba3-8fdfe5b228a5.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for the flight status for the flight 12345678 leaving on april 7\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.658, 0.252, 0.894, 0.282] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_32 | {
"bytes": "",
"path": "./images/82c5913d-8392-44a2-8873-6627a281fa23_3f8b1ce5-738a-4d8e-8fe9-e682ff2cb865.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Calculate monthly payment for car value of $50,000 and downpayment of $10,000 with interest rate of 4% for 60 months.\nPrevious actions:\n[menuitem] Finance -> HOVER"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.376, 0.045, 0.468, 0.068] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_33 | {
"bytes": "",
"path": "./images/91f56f3b-e3de-4d57-95df-f976a11c64f7_589275b2-8649-4dcd-b815-bca201d28836.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a veterinarian emergency service for dogs in Hawaii with a BBB rating of A+\nPrevious actions:\n[link] All -> CLICK\n[label] Veterinarians -> CLICK\n[label] Veterinarian Emergency Services -> CLICK\n[heading] Features -> CLICK\n[label] BBB Rated A+/A -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.429, 0.248, 0.559, 0.265] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_34 | {
"bytes": "",
"path": "./images/01bede1e-91ec-4fe0-a6bc-173826497800_c027c8ec-b3e0-44d0-b671-5700374e6284.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a person by phone number 123456789.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.188, 0.211, 0.26, 0.223] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_35 | {
"bytes": "",
"path": "./images/d5054276-2223-44f3-b2ed-4944bae4d2b1_1113a688-3969-4e5b-9a16-d418ef8ac466.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a Hotel in Boston zip code 02199 and make a pick-up reservation in a suv with gps for 2 people on March 27 11 am and return on March 29, 1 pm.\nPrevious actions:\n[button] Selected Pick-Up Date 03/23/2023 -> CLICK\n[button] 03/27/2023 -> CLICK\n[combobox] Pick-Up Time Selector -> SELECT: 11:00 AM\n[button] Date -> CLICK\n[button] 03/29/2023 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.514, 0.178, 0.619, 0.211] in the image\nAnd my action is Action: SELECT\nValue: 1 00 PM"
}
] |
mind2web_36 | {
"bytes": "",
"path": "./images/ca049641-9721-4593-95c5-a47e22365b5b_fd1a35c4-45cd-4e46-ba38-ceb3203b6cb9.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the cheapest new electric car with highest ev mile range per charge, see the details of the top result.\nPrevious actions:\n[checkbox] New (9,776) -> CLICK\n[heading] Fuel Type -> CLICK\n[checkbox] Electric (175) -> CLICK\n[heading] Electric Mile Range -> CLICK\n[checkbox] 300+ Miles (12) -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.775, 0.159, 0.888, 0.182] in the image\nAnd my action is Action: SELECT\nValue: Price - Lowest"
}
] |
mind2web_37 | {
"bytes": "",
"path": "./images/db203a3a-8e62-41aa-ac05-717a8c11508d_3f1ad2d8-508b-4dca-b072-8d2ff125fafe.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Plan a trip from Brooklyn, NY to Staten Island, NY on March 25th mainly using bus and subway terminal. The trip should arive at 9:45 AM.\nPrevious actions:\n[select] 1 -> SELECT: 9\n[select] 00 -> SELECT: 45\n[select] AM -> SELECT: AM\n[button] 3/25/2023 -> CLICK\n[link] 25, Saturday March 2023 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.141, 0.449, 0.359, 0.476] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_38 | {
"bytes": "",
"path": "./images/504c0c6b-7e78-4bfa-ae3f-00f8e59c3693_9abb53ba-45cd-4c7f-92ee-33073e99789b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a parking lot in Gloucester and book a ride from there to North Plymouth on April 28, 2:30 pm, view the map to understand the route better.\nPrevious actions:\n[button] Transit -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.379, 0.102, 0.621, 0.118] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_39 | {
"bytes": "",
"path": "./images/b2c18588-b115-4937-b69f-8feeab22c387_b32b39aa-4510-4ee1-8d3f-560a4fb3220f.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Locate the list of movies \"at home\", sorted by most recent.\nPrevious actions:\n[link] MOVIES -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.391, 0.207, 0.505, 0.239] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_40 | {
"bytes": "",
"path": "./images/aab91310-0100-4a40-98e9-720c53199bff_600552f9-d248-4c02-bede-2b4624a229da.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find mens hiking shoes under $100 in size 10 that has biggest discount.\nPrevious actions:\n[textbox] Search by keyword or web id -> TYPE: mens hiking shoes\n[input] -> CLICK\n[div] Price -> CLICK\n[link] $10 to $25 (3) -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.016, 0.501, 0.109, 0.514] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_41 | {
"bytes": "",
"path": "./images/ddee9314-5a29-4258-b11b-b6432ec719f1_ad2fa916-8d5f-41f8-bd96-ea4924d38c52.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for all free virtual business start-up events happening tomorrow in Las Vegas and follow the organizer with most followers.\nPrevious actions:\n[link] Business -> CLICK\n[svg] -> CLICK\n[combobox] autocomplete -> TYPE: LAS VEGAS\n[div] Las Vegas -> CLICK\n[svg] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.074, 0.342, 0.254, 0.373] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_42 | {
"bytes": "",
"path": "./images/40cd58cd-6c9a-47b9-a927-92243970d87a_5f3ff4a9-f9ef-4e5b-99de-5bbf85c5f02e.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find monthly daytime only parking nearest to Madison Square Garden starting from April 22\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.476, 0.076, 0.523, 0.09] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_43 | {
"bytes": "",
"path": "./images/1d0dcb84-da65-4830-ba60-cc2b09a52a13_3f6cc639-ac20-4823-b0b9-b6bb1a1c9d26.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find plastic bathroom baskets that are gray from low to high prices near zip code 60173.\nPrevious actions:\n[link] Products -> CLICK\n[button] Bathroom -> CLICK\n[link] Bathroom storage -> CLICK\n[searchbox] Search by ZIP code or city, state -> TYPE: 60173\n[button] Submit -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.863, 0.269, 0.981, 0.296] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_44 | {
"bytes": "",
"path": "./images/08a0c5cc-41a9-4eb8-b3ea-91d9e26f8bd3_9e62b475-81ab-4342-974b-bc13968dad2c.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get directions for the Koa resort campground in Arizona\nPrevious actions:\n[link] CAMPING AT KOA -> CLICK\n[link] KOA RESORT CAMPGROUNDS -> CLICK\n[button] Arizona -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.165, 0.249, 0.35, 0.26] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_45 | {
"bytes": "",
"path": "./images/97e3f951-891d-4626-8dbb-ab6e39261d05_4dd5c447-4344-47c2-aaf4-63554db508f5.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find first class flight for May 27 from JFK, New York to Heathrow, London for 2 adults, one 4 year old child and one infant under 1 year old who will sit on a lap.\nPrevious actions:\n[tab] Flights -> CLICK\n[tab] One-way -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.787, 0.092, 0.85, 0.107] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_46 | {
"bytes": "",
"path": "./images/4775a0e3-3dad-46f2-b719-b1c0dae9e147_cff52937-7d1f-4306-8a48-62e7d8b814fa.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find tortillas bread in grocery stock near zip 59901 and add 5 pieces of 12 pcs pack in the cart and view the cart.\nPrevious actions:\n[button] COLUMBUS, OH -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.198, 0.077, 0.294, 0.089] in the image\nAnd my action is Action: TYPE\nValue: 59901"
}
] |
mind2web_47 | {
"bytes": "",
"path": "./images/4a0bd619-4aa9-48d8-8322-89a86aef93db_33ee7882-a48b-49b5-afd7-d34ebec0a600.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a newly listed used antique Oak chair in the French style made before year 1800.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.267, 0.071, 0.321, 0.08] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_48 | {
"bytes": "",
"path": "./images/8b6eed27-39b4-4dbc-97d3-eeec13a8817d_e17eb0b6-cf6e-45af-be58-704816835353.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book a roundtrip on July 1 from Mumbai to London and vice versa on July 5 for two adults and a 12-year-old in premium economy and if a flight is not available, search through the calendar for available flights.\nPrevious actions:\n[button] Wednesday, July 5, 2023 -> CLICK\n[button] 1 Adult -> CLICK\n[button] Substract one Adult -> CLICK\n[button] Substract one Child -> CLICK\n[generic] Economy -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.076, 0.207, 0.271, 0.241] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_49 | {
"bytes": "",
"path": "./images/2705de3e-4461-4668-8573-8f10c1eb6641_2fa1df82-2bfc-4e2a-a8dc-bf00f7ea75a4.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a low-priced refurbished Microsoft Xbox one console in white color and available in stock.\nPrevious actions:\n[button] Main Menu -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.0, 0.279, 0.291, 0.321] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_50 | {
"bytes": "",
"path": "./images/92b51ef3-7b37-4423-aa6b-e8fb6bf32156_d4ad5462-08d1-400f-881b-080390e948c2.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: search for news about the latest news about Lebron James' injury and share the article on twitter.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.219, 0.025, 0.728, 0.046] in the image\nAnd my action is Action: TYPE\nValue: lebron james"
}
] |
mind2web_51 | {
"bytes": "",
"path": "./images/6b4aa5a9-79c5-49a3-9cc6-ead02159a3dd_abcaa74b-7460-4b0b-95e6-3fcf23ac1904.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a Queen size memory foam mattress with lowest price.\nPrevious actions:\n[button] Menu -> CLICK\n[button] Products -> CLICK\n[button] Beds & mattresses -> CLICK\n[link] Mattresses -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.202, 0.333, 0.298, 0.398] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_52 | {
"bytes": "",
"path": "./images/a9708ad7-bec8-4435-b055-8ef6da520b8c_332d0daa-f81f-45b6-aa45-2bb32665819c.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Play the top track for the top indie artist in the last 30 days\nPrevious actions:\n[link] Music -> CLICK\n[link] indie -> CLICK\n[link] The Smiths -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.578, 0.454, 0.645, 0.464] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_53 | {
"bytes": "",
"path": "./images/edf748d4-07cd-4f0b-aad3-01baebbd557b_0d5293b2-9ff7-48b5-80f6-b043d52c9066.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book the cheapest available hotel for a three night stay from 1st June in Jakarta. The guest is named Joe Bloggs with the email address of buckeye.foobar@gmail.com and phone number of 11111111111. Billing address is in New York, zip code 10001\n.\nPrevious actions:\n[div] Jun -> CLICK\n[generic] 1 -> CLICK\n[generic] 4 -> CLICK\n[button] Search -> CLICK\n[button] Yes, I agree -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.298, 0.077, 0.386, 0.106] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_54 | {
"bytes": "",
"path": "./images/05c4da5b-263d-40a4-9982-6cf6311b57a1_770aa60a-7be7-436b-9b5d-59111c135246.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find an Xbox Wireless controller rated above 4 stars.\nPrevious actions:\n[searchbox] Search Site -> TYPE: xbox wireless controller\n[button] -> CLICK\n[dt] Customer Ratings -> CLICK\n[div] & up -> CLICK\n[button] APPLY -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.248, 0.369, 0.451, 0.526] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_55 | {
"bytes": "",
"path": "./images/f804150b-c2e9-40ca-b71f-d0d0d95e0c46_0b2e5b9a-bed8-4064-8057-bb32b4bc6111.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the cheapest hotel in New York City for 2 adults from Apr7 to Apr 8.\nPrevious actions:\n[link] Hotel , Opens another site in a new window that ma... -> CLICK\n[searchbox] Type your destination -> TYPE: New York City\n[option] New York (New York City) New York State, United St... -> CLICK\n[checkbox] 7 April 2023 -> CLICK\n[checkbox] 8 April 2023 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.82, 0.125, 0.927, 0.157] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_56 | {
"bytes": "",
"path": "./images/b48c9974-4ba0-4112-98ce-3667781fa71b_459ed167-c817-4bbf-bc91-73822e98bfd9.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a cheapest standard train ticket from London to Sheffield on March, 29, arriving by 10:45 am for 2 adults.\nPrevious actions:\n[combobox] Departure station, none selected. Select a station... -> TYPE: london\n[span] London -> CLICK\n[combobox] Arrival station, none selected. Select a station. -> TYPE: sheffield"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.114, 0.14, 0.326, 0.166] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_57 | {
"bytes": "",
"path": "./images/725c38e7-7987-40b3-8f19-8450f8277f06_b75be239-fb3b-4d79-820c-e374efbe2c73.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find the cheapest event to attend in Athens, Greece on April 2nd\nPrevious actions:\n[input] -> CLICK\n[button] Greece -> CLICK\n[button] Athens -> CLICK\n[button] Go! -> CLICK\n[div] Sort by -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.572, 0.123, 0.689, 0.136] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_58 | {
"bytes": "",
"path": "./images/af97084c-5b72-4fec-be1e-dcab0980b335_251534f2-9acc-4f2c-a1d4-2158f8a4840e.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Check all Bayern Munich goalie stats\nPrevious actions:\n[button] MORE -> CLICK\n[div] Soccer -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.181, 0.136, 0.468, 0.17] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_59 | {
"bytes": "",
"path": "./images/5d69e640-7765-4381-979c-9881afc048dd_b9ea57b4-ce6f-4010-b79b-f3f8fc031d1b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Calculate the monthly payment of a 15k car with a down payment of 5k in Michigan for a fair-rated person for a term length of 48 months, then shop with this budget for the cheapest car and add it to the favorites.\nPrevious actions:\n[link] Finance -> CLICK\n[link] Car payment calculator -> CLICK\n[textbox] Vehicle Price -> TYPE: 15000"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.159, 0.23, 0.459, 0.253] in the image\nAnd my action is Action: TYPE\nValue: 5000"
}
] |
mind2web_60 | {
"bytes": "",
"path": "./images/b4362dec-6803-415a-a112-819f694b84d9_3a37f05f-1cf6-43cb-9509-7936404dae33.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Purchase a one meal season dining pass for the Six flags great america\nPrevious actions:\n[span] Six Flags Great America -> CLICK\n[button] Go! -> CLICK\n[link] Tickets & Passes -> CLICK\n[link] One Day Add-Ons -> CLICK\n[link] Buy Now -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.425, 0.0, 0.505, 0.025] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_61 | {
"bytes": "",
"path": "./images/10e17667-f9cf-4a68-adcd-ad2677c3f385_a87e7411-3bda-4944-beb8-2f77f9fbe4b5.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find series 1 and series 2 BMW vehicles in the store nearest to 07055\nPrevious actions:\n[link] SEARCH CARS -> CLICK\n[div] Best match -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.038, 0.288, 0.243, 0.297] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_62 | {
"bytes": "",
"path": "./images/504c0c6b-7e78-4bfa-ae3f-00f8e59c3693_bbff6535-c2fa-4fe3-ab52-3ba6813014b0.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a parking lot in Gloucester and book a ride from there to North Plymouth on April 28, 2:30 pm, view the map to understand the route better.\nPrevious actions:\n[button] Depart at 8:20 AM, 4/10/23 -> CLICK\n[LabelText] Monday, April 10, 2023, click or press the enter o... -> CLICK\n[gridcell] Friday, April 28, 2023 -> CLICK\n[combobox] Hour -> TYPE: 2\n[combobox] Minute -> TYPE: 30"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.254, 0.424, 0.305, 0.457] in the image\nAnd my action is Action: SELECT\nValue: PM"
}
] |
mind2web_63 | {
"bytes": "",
"path": "./images/cf361c84-6414-4b05-a7a1-77383997150a_3fed27c4-2cb3-43d0-b92a-16275e1f8178.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get an SUV with an additional driver and wifi for pick up in any rental location near Washington regional airport on June 1, 11 am, and drop off at Washington international airport on June 2, 11 am, and pay for the booking instantly.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.84, 0.02, 0.93, 0.03] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_64 | {
"bytes": "",
"path": "./images/1860601a-faef-4dba-8cbb-807ea8434dca_894a3e88-d3f2-417d-b464-ce6f3086c9cd.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Check for travel requirements when travelling from New York to Tokyo as US citizen.\nPrevious actions:\n[link] Travel Planning Center -> CLICK\n[button] Find Your Destination -> CLICK\n[link] Explore Destinations & Travel Requirements -> CLICK\n[combobox] Origin -> CLICK\n[combobox] Origin -> TYPE: New York"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.068, 0.324, 0.236, 0.353] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_65 | {
"bytes": "",
"path": "./images/04ec089f-2ae1-4fa9-bfdf-4dff9ab710ad_be4c997b-89fc-4e4f-93d1-092dc7cde1a4.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find highest rated white water rafting experience available in Montana from May 1 to May 7\nPrevious actions:\n[circle] -> CLICK\n[button] -> CLICK\n[button] -> CLICK\n[div] White Water Rafting -> CLICK\n[span] Featured -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.82, 0.195, 0.905, 0.203] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_66 | {
"bytes": "",
"path": "./images/b3c7e28e-00a2-4ef7-89c5-f3ad7efadd6a_409090be-7df9-412d-b354-2a68656eb421.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Show my Library and sort by Albums.\nPrevious actions:\n[img] james9091 -> CLICK\n[link] Library -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.205, 0.301, 0.285, 0.328] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_67 | {
"bytes": "",
"path": "./images/6d5c4d19-c2cc-4e68-9f51-71ba6cd6e38a_d905eacd-ba9e-43ac-815c-d4e42a636301.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a hotel with 4 star and above rating in Los Angeles for 3 days next week and save it.\nPrevious actions:\n[link] Search for hotels -> CLICK\n[textbox] Enter a city, hotel, airport, address or landmark -> TYPE: Los Angeles"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.223, 0.117, 0.573, 0.149] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_68 | {
"bytes": "",
"path": "./images/f86b0a14-ed91-491d-85be-b0af0b849353_f7f63924-6669-400e-b187-76d3b6243151.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse movies in theaters that are certified fresh and sorty by highest audience score.\nPrevious actions:\n[link] MOVIES -> CLICK\n[svg] -> CLICK\n[div] -> CLICK\n[button] APPLY -> CLICK\n[svg] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.34, 0.447, 0.36, 0.461] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_69 | {
"bytes": "",
"path": "./images/1a16a98f-4e8f-4ff9-9fa2-b97d2d403cc3_a824cebd-8374-4f3a-b76d-df0f6a9f45ae.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Add a set of sonoma bath towels to the cart and apply a coupon code for free shipping.\nPrevious actions:\n[textbox] Search by keyword or web id -> TYPE: bath towels\n[span] bath towels -> CLICK\n[img] Sonoma Goods For Life® Ultimate Bath Towel with Hy... -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.67, 0.335, 0.759, 0.355] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_70 | {
"bytes": "",
"path": "./images/e4f8a347-b288-4a00-9e6b-89cbccda42d2_f73c9f1b-6d77-4f34-bbdc-84ab137a91f6.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for a parking near any airport in california starting april 20 and ending april 23\nPrevious actions:\n[link] View All Airports -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.75, 0.296, 0.831, 0.312] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_71 | {
"bytes": "",
"path": "./images/4b2030ff-b83c-445f-bf87-9c8fbc68498b_ff0ee393-4490-4aaf-9fc0-a21fcdb41c9b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse for wall art with a price range of $25 to $50.\nPrevious actions:\n[textbox] Search by keyword or web id -> TYPE: wall art"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.168, 0.027, 0.187, 0.036] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_72 | {
"bytes": "",
"path": "./images/998e55fb-66d7-443f-bb73-9cb6a2bc26a8_47baddf3-e09d-414f-8c3a-7de89a39aa06.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find used BMW i3 for 10001 zip code with lowest price.\nPrevious actions:\n[menuitem] Buy -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.012, 0.206, 0.324, 0.225] in the image\nAnd my action is Action: SELECT\nValue: BMW"
}
] |
mind2web_73 | {
"bytes": "",
"path": "./images/754c8c29-92b6-4af1-9594-f9d3da37b951_7b44e78e-d6d9-44ad-8331-3930b3d959b6.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a 3-star rated hostel with balcony and free cancellation option in Udipi, Karnataka for one person under 50 dollars from June 15 to 20, free wifi is compulsory.\nPrevious actions:\n[link] Hostels -> CLICK\n[searchbox] Please type your destination -> TYPE: udupi"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.073, 0.173, 0.409, 0.195] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_74 | {
"bytes": "",
"path": "./images/a11022ab-f733-4295-a2f2-0da19cccc1b4_6ba16278-a7a3-4e06-8426-356b05e3219c.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Sign up for emails using the name Dick Smith and the email smith@gmail.com, select Six Flags Magic Mountain as the Park of Preference,\nPrevious actions:\n[textbox] First Name -> TYPE: Dick\n[textbox] Last Name -> TYPE: Smith"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.322, 0.258, 0.678, 0.326] in the image\nAnd my action is Action: TYPE\nValue: smith@gmail.com"
}
] |
mind2web_75 | {
"bytes": "",
"path": "./images/bf008019-bffd-42ad-a48b-054488e1458c_98ce6733-1dd3-4cf0-a29c-03f67319dc68.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a parking near any NBA arena to enter today after 5 pm and exit after 8 pm\nPrevious actions:\n[link] View All Stadiums -> CLICK\n[link] Book Now -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.261, 0.251, 0.314, 0.274] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_76 | {
"bytes": "",
"path": "./images/f57e6c0a-8f8b-4756-9f1d-1bdea7a0af5c_276ea5ec-d119-4d03-9121-f9ee4616da2a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Add a pack of toilet paper and a bottle of laundry detergent to your Amazon cart with the lowest total price.\nPrevious actions:\n[textbox] Search Amazon -> TYPE: toilet paper\n[button] Go -> CLICK\n[span] Featured -> CLICK\n[option] Price: Low to High -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.462, 0.607, 0.53, 0.623] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_77 | {
"bytes": "",
"path": "./images/f4dcbd09-45f3-47b5-a49d-36658d5b50ca_9a924659-4b0a-4374-9e9c-e24889c4dac5.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find tickets for Pizza making class in New York next month.\nPrevious actions:\n[svg] -> CLICK\n[svg] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.281, 0.071, 0.719, 0.098] in the image\nAnd my action is Action: TYPE\nValue: NEW YORK"
}
] |
mind2web_78 | {
"bytes": "",
"path": "./images/7a698566-2a8f-4d9d-9da9-17288b66917f_3c3caece-89d2-4110-b976-242c6070e947.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find hotels in Tokyo for two adults and one child, with a budget of $500 per night for 2 nights starting on May 5th.\nPrevious actions:\n[combobox] Enter a destination or property -> TYPE: Tokyo\n[generic] Japan -> CLICK\n[div] 5 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.73, 0.329, 0.736, 0.345] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_79 | {
"bytes": "",
"path": "./images/d538537c-ff96-4918-8807-af09b26199d1_794b3de4-0e28-4ba9-819a-017558734d98.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a flights between SFO and EWR using the Boeing 787-9 aircraft\nPrevious actions:\n[textbox] Destination -> CLICK\n[textbox] Destination -> TYPE: EWR\n[div] Newark Liberty Intl (Newark) - -> CLICK\n[img] Submit Search -> CLICK\n[span] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.188, 0.363, 0.205, 0.371] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_80 | {
"bytes": "",
"path": "./images/73cf6eec-cae6-4d5b-9b8e-e44359311565_702d51c4-4747-4525-b58c-324c776f600a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Sign up for the newsletter\nPrevious actions:\n[textbox] Email Address * -> TYPE: larryknox@gmail.com"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.292, 0.685, 0.337, 0.718] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_81 | {
"bytes": "",
"path": "./images/4b2bb9e4-8e28-490b-895f-80da5e7e2815_c3c9c397-7337-484c-97c1-71421f964f62.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get email alerts for a flight from BHZ to EWN leaving on april 30 on the email lin.lon@gmail.com\nPrevious actions:\n[link] 30 -> CLICK\n[button] Search -> CLICK\n[button] Get alerts for this flight for flight 906 American... -> CLICK\n[textbox] Email -> TYPE: lin.lon@gmail.com\n[textbox] Confirm email -> TYPE: lin.lon@gmail.com"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.645, 0.387, 0.83, 0.412] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_82 | {
"bytes": "",
"path": "./images/21e5c264-df85-4055-a566-ecb65cdd8c63_4c93db0b-982d-4815-933d-10283c2fb380.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Add to my basket tickets for 3 people from New York to Alfred.\nPrevious actions:\n[span] Alfred, NY -> CLICK\n[textbox] Pick a date -> CLICK\n[gridcell] March 31, 2023 -> CLICK\n[span] -> CLICK\n[span] -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.763, 0.203, 0.831, 0.217] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_83 | {
"bytes": "",
"path": "./images/1684f224-47d5-45ed-9a26-821b98b851cd_eba8a877-10db-482e-8ca8-f6b61efb119b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Look for the highest reviewed for a 4-adults stay in Hollywood, USA, on April 19 to 23 with an option for free cancellation with a price range between 150 to 200 dollars.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.195, 0.529, 0.829, 0.545] in the image\nAnd my action is Action: TYPE\nValue: HOLLYWOOD"
}
] |
mind2web_84 | {
"bytes": "",
"path": "./images/40cd58cd-6c9a-47b9-a927-92243970d87a_50d4a729-d12a-4707-af8d-69b5ab13c8db.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find monthly daytime only parking nearest to Madison Square Garden starting from April 22\nPrevious actions:\n[tab] Monthly -> CLICK\n[textbox] Search for parking -> TYPE: madison square garden\n[li] Madison Square Garden, Pennsylvania Plaza, New Yor... -> CLICK\n[select] All Parking Options -> SELECT: Daytime Only Parking\n[textbox] Monthly Start Date -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.207, 0.491, 0.219, 0.506] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_85 | {
"bytes": "",
"path": "./images/cf89b0ae-39ca-4451-80df-260d46c62f21_0043be09-27a3-4b47-81c8-cc4ee1cb996e.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a hotel in Ohio From July 2nd to July 8th for 2 adults.\nPrevious actions:\n[button] Next month -> CLICK\n[button] Next month -> CLICK\n[button] Next month -> CLICK\n[button] Jul 2, 2023 -> CLICK\n[button] Jul 8, 2023 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.929, 0.357, 0.984, 0.377] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_86 | {
"bytes": "",
"path": "./images/5e01c2f7-c63b-4826-98a3-a18f731e16ea_9a4dabad-f38c-4fea-9345-202450e96322.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse tickets for Chicago Bears games.\nPrevious actions:\n[button] SPORTS -> HOVER\n[tab] NFL -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.342, 0.149, 0.455, 0.162] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_87 | {
"bytes": "",
"path": "./images/27a8bf82-2e7e-48c3-8fb1-66909c13d585_66ece2a6-3789-462d-8cd2-627355cd988a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a Buick for sale within 250 miles from 26807.\nPrevious actions:\n[combobox] All Makes -> CLICK\n[textbox] Zip Code: 43085 -> TYPE: 26807\n[combobox] 25 miles -> CLICK\n[generic] 250 miles -> CLICK\n[button] View 95 Vehicles -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.038, 0.357, 0.227, 0.37] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_88 | {
"bytes": "",
"path": "./images/4d73937b-3be7-4f5f-950f-b1905244a2ac_cf8a0b6f-dde7-4ec5-959a-23c1fd331528.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Show MLB tickets for this weekend and select the next one.\nPrevious actions:\n[button] SPORTS -> HOVER\n[link] All MLB Tickets -> CLICK\n[button] All dates -> CLICK\n[link] This weekend -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.864, 0.239, 0.941, 0.266] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_89 | {
"bytes": "",
"path": "./images/d9e9c178-e945-49fe-9d06-a01cf02ab2db_98b66f05-3799-48a7-955d-5c2075f75a44.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Browse the highest rated spas that are $$ dollar signs and under.\nPrevious actions:\n[button] Price: -> CLICK\n[input] -> CLICK\n[input] -> CLICK\n[span] Save -> CLICK\n[button] Recommended -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.518, 0.106, 0.612, 0.122] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_90 | {
"bytes": "",
"path": "./images/58f811fd-0f17-430b-befc-885605e13e41_1c9cabc3-3fc6-4561-891d-5b473220ab9b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Cancel an order with order number X123456 of Harry Potter Box Set as not available at address.\nPrevious actions:\n[link] Cancel my order -> CLICK\n[input] -> TYPE: James Smith\n[textbox] Email address used to place your order * -> TYPE: buckeye.foobar@gmail.com\n[textbox] Order number * -> TYPE: X123456"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.235, 0.43, 0.759, 0.457] in the image\nAnd my action is Action: TYPE\nValue: 08/04/23"
}
] |
mind2web_91 | {
"bytes": "",
"path": "./images/9326b908-cbe6-41f6-957f-00b84c26bfcc_9d73a27e-5499-4d8d-84c2-f442fdfd516e.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find recently added guitar pro non-acoustic tab intros\nPrevious actions:\n[link] Tabs -> CLICK\n[link] Non-acoustic 1,805,658 -> CLICK\n[link] Intro 65,171 -> CLICK\n[link] Guitar Pro -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.824, 0.204, 0.97, 0.216] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_92 | {
"bytes": "",
"path": "./images/dd13bc51-f582-4004-8641-eb4e62cabfc7_25d69ce3-0527-4b54-ae05-76b4246c6816.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Get prequalified with a good credit score for a $30000 car with a $6000 down payment for 48 months.\nPrevious actions:\n[menuitem] Finance -> HOVER"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.213, 0.046, 0.358, 0.069] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_93 | {
"bytes": "",
"path": "./images/1ee63f83-8b6a-4883-813f-63f589e6e52b_c4186aed-2d4d-41ca-bc89-ec1e003fc4b7.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Book two rooms in the cheapest 3-star hotel which provides free breakfast, and free airport transportation for three adults, and one child from April 22 to 25 in texas city, the hotel should provide amenities for kids.\nPrevious actions:\n[button] sub 1 -> CLICK\n[div] open -> CLICK\n[option] 6 -> CLICK\n[button] Update -> CLICK\n[button] Search -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.461, 0.449, 0.606, 0.476] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_94 | {
"bytes": "",
"path": "./images/2a7f7732-ec9b-4d78-91ad-f0fe10bf7daf_9eb3684f-bdc3-44d7-aa67-be7839fb83ec.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Find a gaming pc which can run Call of Duty Warzone at 4k.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.016, 0.013, 0.05, 0.044] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_95 | {
"bytes": "",
"path": "./images/0dd6fbb7-4236-47f7-a69c-30f0836b76c3_16a62e54-3ff2-4ec8-aa34-0f75f384d352.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: See the price trends of used Tesla cars from August 1, 2022, to January 31, 2023, and compare the trend with the Cargurus index.\nPrevious actions:\n[span] -> CLICK\n[generic] Previous Month -> CLICK\n[generic] Previous Month -> CLICK\n[generic] Previous Month -> CLICK\n[gridcell] 31 -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.096, 0.199, 0.182, 0.222] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_96 | {
"bytes": "",
"path": "./images/8e721b00-f406-449b-9885-0267b47ecfdb_cd58814d-9500-4922-a7f4-416a19ffcc34.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Search for used Jaguar XFs with no black exterior color and save the search as Jaguar to get a notification daily.\nPrevious actions:\n[searchbox] Search make, model, or keyword -> TYPE: jaguar xf\n[link] Jaguar XF -> CLICK\n[button] Advanced Search -> CLICK\n[textbox] Ex. Black exterior -> TYPE: black\n[button] Black (Exterior Color) -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.069, 0.16, 0.168, 0.173] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_97 | {
"bytes": "",
"path": "./images/4f75897b-8df4-4f4c-88c3-72d440f6a4ee_06b3b17f-68a0-4d62-a236-9852c9ae658a.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Show me all the destinations in Armenia, Austria, Belgium and Bulgaria in Alphabetical order\nPrevious actions:\n[label] Austria -> CLICK\n[label] Belgium -> CLICK\n[label] Bulgaria -> CLICK\n[div] Popular -> CLICK\n[div] A - Z -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.051, 0.173, 0.263, 0.195] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_98 | {
"bytes": "",
"path": "./images/8b130a70-f03a-4f86-8f2a-44f1b44f3655_3533eae9-5554-4489-9498-64ba4f8c832b.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Check my trip for Smith and booking number X899987799.\nPrevious actions:\nNone"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.342, 0.397, 0.657, 0.447] in the image\nAnd my action is Action: CLICK"
}
] |
mind2web_99 | {
"bytes": "",
"path": "./images/3596733f-6427-4ab5-9ff1-c65baaf524f1_0a0d2fc8-04cf-49c2-a658-cf12ef65c5eb.png"
} | [
{
"from": "human",
"value": "<image>\nImagine that you are imitating humans doing web navigation for a task step by step. At each stage, you can see the webpage like humans by a screenshot and know the previous actions before the current step decided by yourself through recorded history. You need to decide on the first following action to take. You can click an element with the mouse, select an option, or type text with the keyboard. (For your understanding, they are like the click(), select_option() and type() functions in playwright respectively) One next step means one operation within the three.\nPlease pinpoint the bounding box coordinates of the element’s region that I should click in order to complete the following instruction: Show the list of restaurants with reservations available for 04/11 with vegetarian food.\nPrevious actions:\n[textbox] Search restaurants, cuisines, etc. -> TYPE: vegetarian\n[b] Vegetarian -> CLICK"
},
{
"from": "gpt",
"value": "For my next step, the target element of my action is bounded by the box [0.016, 0.22, 0.075, 0.231] in the image\nAnd my action is Action: CLICK"
}
] |
End of preview. Expand
in Dataset Viewer.
Mind2Web training set for the paper: Harnessing Webpage Uis For Text Rich Visual Understanding
🌐 Homepage | 🐍 GitHub | 📖 arXiv
Introduction
We introduce MultiUI, a dataset containing 7.3 million samples from 1 million websites, covering diverse multi- modal tasks and UI layouts. Models trained on MultiUI not only excel in web UI tasks—achieving up to a 48% improvement on VisualWebBench and a 19.1% boost in action accuracy on a web agent dataset Mind2Web—but also generalize surprisingly well to non-web UI tasks and even to non-UI domains, such as document understanding, OCR, and chart interpretation.
Contact
- Junpeng Liu: jpliu@link.cuhk.edu.hk
- Xiang Yue: xyue2@andrew.cmu.edu
Citation
If you find this work helpful, please cite out paper:
@misc{liu2024harnessingwebpageuistextrich,
title={Harnessing Webpage UIs for Text-Rich Visual Understanding},
author={Junpeng Liu and Tianyue Ou and Yifan Song and Yuxiao Qu and Wai Lam and Chenyan Xiong and Wenhu Chen and Graham Neubig and Xiang Yue},
year={2024},
eprint={2410.13824},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2410.13824},
}
- Downloads last month
- 616