{
"cells": [
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"c:\\Users\\Jose\\Desktop\\Nuanced_Recommendation_System\\.venv\\Lib\\site-packages\\transformers\\tokenization_utils_base.py:1617: FutureWarning: `clean_up_tokenization_spaces` was not set. It will be set to `True` by default. This behavior will be deprecated in transformers v4.45, and will be then set to `False` by default. For more details check this issue: https://github.com/huggingface/transformers/issues/31884\n",
" warnings.warn(\n"
]
}
],
"source": [
"import os\n",
"import json\n",
"\n",
"from sentence_transformers import SentenceTransformer\n",
"from sklearn.metrics.pairwise import cosine_similarity\n",
"\n",
"model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')\n",
"\n",
"import gradio as gr"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"def get_n_weighted_scores(embeddings, query, n, objective_weight, subjective_weight):\n",
" query = [model.encode(query)]\n",
"\n",
" weighted_scores = []\n",
"\n",
" for key, value in embeddings.items():\n",
" objective_embedding = value['objective_embedding']\n",
" subjective_embeddings = value['subjective_embeddings']\n",
" \n",
" objective_score = cosine_similarity(query, objective_embedding).item()\n",
" subjective_scores = cosine_similarity(query, subjective_embeddings)\n",
"\n",
" max_score = 0\n",
" max_review_index = 0\n",
" for idx, score in enumerate(subjective_scores[0].tolist()):\n",
" weighted_score = ((objective_score * objective_weight)+(score * subjective_weight))\n",
" if weighted_score > max_score:\n",
" max_score = weighted_score\n",
" max_review_index = idx\n",
" \n",
" weighted_scores.append((key, max_score, max_review_index))\n",
" \n",
" return sorted(weighted_scores, key=lambda x: x[1], reverse=True)[:n]"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"def filter_anime(embeddings, genres, themes, rating):\n",
" genres = set(genres)\n",
" themes = set(themes)\n",
" rating = set(rating)\n",
"\n",
" filtered_anime = embeddings.copy()\n",
" for key, anime in embeddings.items():\n",
"\n",
" anime_genres = set(anime['genres'])\n",
" anime_themes = set(anime['themes'])\n",
" anime_rating = set([anime['rating']])\n",
"\n",
" if genres.intersection(anime_genres) or 'ALL' in genres:\n",
" pass\n",
" else:\n",
" filtered_anime.pop(key)\n",
" continue\n",
" if themes.intersection(anime_themes) or 'ALL' in themes:\n",
" pass\n",
" else:\n",
" filtered_anime.pop(key)\n",
" continue\n",
" if rating.intersection(anime_rating) or 'ALL' in rating:\n",
" pass\n",
" else:\n",
" filtered_anime.pop(key)\n",
" continue\n",
" \n",
" return filtered_anime"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"with open('./embeddings/data.json') as f:\n",
" data = json.load(f)\n",
" embeddings = data['embeddings']\n",
" filters = data['filters']"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [],
"source": [
"def get_recommendation(query, number_of_recommendations, genres, themes, rating, objective_weight, subjective_weight):\n",
" filtered_anime = filter_anime(embeddings, genres, themes, rating)\n",
" results = []\n",
" weighted_scores = get_n_weighted_scores(filtered_anime, query, number_of_recommendations, float(objective_weight), float(subjective_weight))\n",
" for idx, (key, score, review_index) in enumerate(weighted_scores, start=1):\n",
" data = embeddings[key]\n",
" if not data['english']:\n",
" name = data['japanese']\n",
" else:\n",
" name = data['english']\n",
" description = data['description']\n",
" review = data['reviews'][review_index]['text']\n",
" image = data['image']\n",
"\n",
" results.append(gr.Image(label=f\"Recommendation {idx}: {name}\",value=image, height=435, width=500, visible=True))\n",
" results.append(gr.Textbox(label=f\"Synopsis\", value=description, max_lines=7, visible=True))\n",
" results.append(gr.Textbox(label=f\"Most Relevant User Review\",value=review, max_lines=7, visible=True))\n",
"\n",
" for _ in range(10-number_of_recommendations):\n",
" results.append(gr.Image(visible=False))\n",
" results.append(gr.Textbox(visible=False))\n",
" results.append(gr.Textbox(visible=False))\n",
" \n",
" return results"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Running on local URL: http://127.0.0.1:7863\n",
"\n",
"To create a public link, set `share=True` in `launch()`.\n"
]
},
{
"data": {
"text/html": [
""
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"c:\\Users\\Jose\\Desktop\\Nuanced_Recommendation_System\\.venv\\Lib\\site-packages\\gradio\\analytics.py:106: UserWarning: IMPORTANT: You are using gradio version 4.44.1, however version 5.0.1 is available, please upgrade. \n",
"--------\n",
" warnings.warn(\n"
]
}
],
"source": [
"with gr.Blocks(theme=gr.themes.Soft(primary_hue='red')) as demo:\n",
" with gr.Row():\n",
" with gr.Column():\n",
" gr.Markdown(\n",
" '''\n",
" # Welcome to the Nuanced Recommendation System!\n",
" ### This system **combines** both objective (synopsis, episode count, themes) and subjective (user reviews) data, in order to recommend the most approprate anime. Feel free to refine using the **optional** filters below! \n",
" '''\n",
" )\n",
" with gr.Column():\n",
" pass\n",
" \n",
"\n",
" with gr.Row():\n",
" with gr.Column() as input_col:\n",
" query = gr.Textbox(label=\"What are you looking for?\")\n",
" number_of_recommendations = gr.Slider(label= \"# of Recommendations\", minimum=1, maximum=10, value=3, step=1)\n",
" genres = gr.Dropdown(label='Genres',multiselect=True,choices=filters['genres'], value=['ALL'])\n",
" themes = gr.Dropdown(label='Themes',multiselect=True,choices=filters['themes'], value=['ALL'])\n",
" rating = gr.Dropdown(label='Rating',multiselect=True,choices=filters['rating'], value=['ALL'])\n",
" objective_weight = gr.Slider(label= \"Objective Weight\", minimum=0, maximum=1, value=.5, step=.1)\n",
" subjective_weight = gr.Slider(label= \"Subjective Weight\", minimum=0, maximum=1, value=.5, step=.1)\n",
" submit_btn = gr.Button(\"Submit\")\n",
"\n",
" examples = gr.Examples(\n",
" examples=[\n",
" ['A sci-fi anime set in a future where AI and robots have become self-aware', 3, ['Action', 'Sci-Fi', 'Fantasy'], ['ALL'], ['PG-13 - Teens 13 or older'], .8, .2],\n",
" ['An anime where a group of students form a band, and the story focuses on their personal growth and struggles with adulthood', 5, ['ALL'], ['Music'], ['PG-13 - Teens 13 or older', 'R - 17+ (violence & profanity)'], .3, .7],\n",
" ['An anime where the main character starts as a villain but slowly redeems themselves', 3, ['Suspense', 'Action'], ['ALL'], ['PG-13 - Teens 13 or older', 'R - 17+ (violence & profanity)'], .2, .8],\n",
" ],\n",
" inputs=[query, number_of_recommendations, genres, themes, rating, objective_weight, subjective_weight],\n",
" )\n",
"\n",
" outputs = []\n",
" with gr.Column():\n",
" for i in range(10):\n",
" with gr.Row():\n",
" with gr.Column():\n",
" outputs.append(gr.Image(height=435, width=500, visible=False))\n",
" with gr.Column():\n",
" outputs.append(gr.Textbox(max_lines=7, visible=False))\n",
" outputs.append(gr.Textbox(max_lines=7, visible=False))\n",
" \n",
"\n",
" submit_btn.click(\n",
" get_recommendation,\n",
" [query, number_of_recommendations, genres, themes, rating, objective_weight, subjective_weight],\n",
" outputs\n",
" )\n",
"\n",
" demo.launch()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": ".venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.7"
}
},
"nbformat": 4,
"nbformat_minor": 2
}