fix readme
Browse files
README.md
CHANGED
@@ -19,17 +19,24 @@ pretty_name: Analogy Question
|
|
19 |
### Dataset Summary
|
20 |
This dataset contains 5 different word analogy questions used in [Analogy Language Model](https://aclanthology.org/2021.acl-long.280/).
|
21 |
|
22 |
-
|
23 |
-
|
|
|
|
|
24 |
| `sat_full`| -/374 | 5 | 2 | [Turney (2005)](https://arxiv.org/pdf/cs/0508053.pdf) |
|
25 |
| `sat` | 37/337 | 5 | 2 | [Turney (2005)](https://arxiv.org/pdf/cs/0508053.pdf) |
|
26 |
| `u2` | 24/228 | 5,4,3 | 9 | [EnglishForEveryone](https://englishforeveryone.org/Topics/Analogies.html) |
|
27 |
| `u4` | 48/432 | 5,4,3 | 5 | [EnglishForEveryone](https://englishforeveryone.org/Topics/Analogies.html) |
|
28 |
| `google` | 50/500 | 4 | 2 | [Mikolov et al., (2013)](https://www.aclweb.org/anthology/N13-1090.pdf) |
|
29 |
| `bats` | 199/1799 | 4 | 3 | [Gladkova et al., (2016)](https://www.aclweb.org/anthology/N18-2017.pdf) |
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
## Dataset Structure
|
35 |
### Data Instances
|
|
|
19 |
### Dataset Summary
|
20 |
This dataset contains 5 different word analogy questions used in [Analogy Language Model](https://aclanthology.org/2021.acl-long.280/).
|
21 |
|
22 |
+
- original analogy questions
|
23 |
+
|
24 |
+
| name | Size (valid/test) | Num of choice | Num of relation group | Original Reference |
|
25 |
+
|-----------|------------------:|--------------:|----------------------:|:--------------------------------------------------------------------------:|
|
26 |
| `sat_full`| -/374 | 5 | 2 | [Turney (2005)](https://arxiv.org/pdf/cs/0508053.pdf) |
|
27 |
| `sat` | 37/337 | 5 | 2 | [Turney (2005)](https://arxiv.org/pdf/cs/0508053.pdf) |
|
28 |
| `u2` | 24/228 | 5,4,3 | 9 | [EnglishForEveryone](https://englishforeveryone.org/Topics/Analogies.html) |
|
29 |
| `u4` | 48/432 | 5,4,3 | 5 | [EnglishForEveryone](https://englishforeveryone.org/Topics/Analogies.html) |
|
30 |
| `google` | 50/500 | 4 | 2 | [Mikolov et al., (2013)](https://www.aclweb.org/anthology/N13-1090.pdf) |
|
31 |
| `bats` | 199/1799 | 4 | 3 | [Gladkova et al., (2016)](https://www.aclweb.org/anthology/N18-2017.pdf) |
|
32 |
+
|
33 |
+
- extra analogy questions
|
34 |
+
|
35 |
+
| name | Size (valid/test) | Num of choice (valid/test) | Num of relation group (valid/test) | Original Reference |
|
36 |
+
|:------------------------------------|:--------------------|:-----------------------------|:-------------------------------------|:-----------------------------------------------------------------------------------------------------------------------|
|
37 |
+
| `semeval2012_relational_similarity` | 79/- | 3/- | 79/- | [relbert/semeval2012_relational_similarity](https://huggingface.co/datasets/relbert/semeval2012_relational_similarity) |
|
38 |
+
| `t_rex_relational_similarity` | 496/183 | 74/48 | 60/19 | [relbert/t_rex_relational_similarity](https://huggingface.co/datasets/relbert/t_rex_relational_similarity) |
|
39 |
+
| `conceptnet_relational_similarity` | 1112/1192 | 19/17 | 18/16 | [relbert/conceptnet_relational_similarity](https://huggingface.co/datasets/relbert/conceptnet_relational_similarity) |
|
40 |
|
41 |
## Dataset Structure
|
42 |
### Data Instances
|
add_new_analogy.py
CHANGED
@@ -1,94 +1,113 @@
|
|
1 |
import json
|
2 |
import os
|
3 |
-
from itertools import combinations
|
4 |
from random import shuffle, seed
|
|
|
|
|
5 |
from datasets import load_dataset
|
6 |
|
7 |
-
# # create analogy from `relbert/semeval2012_relational_similarity`
|
8 |
-
# data = load_dataset("relbert/semeval2012_relational_similarity", split="validation")
|
9 |
-
# analogy_data = [{
|
10 |
-
# "stem": i['positives'][0], "choice": i["negatives"] + [i['positives'][1]], "answer": 2, "prefix": i["relation_type"]
|
11 |
-
# } for i in data]
|
12 |
-
# os.makedirs("dataset/semeval2012_relational_similarity", exist_ok=True)
|
13 |
-
# with open("dataset/semeval2012_relational_similarity/valid.jsonl", "w") as f:
|
14 |
-
# f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
-
|
18 |
-
#
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
# shuffle(negative)
|
35 |
-
# analogy_data.append({
|
36 |
-
# "stem": q.tolist(), "choice": [c.tolist()] + negative[:5], "answer": 0, "prefix": i["relation_type"]
|
37 |
-
# })
|
38 |
-
# os.makedirs("dataset/t_rex_relational_similarity", exist_ok=True)
|
39 |
-
# with open("dataset/t_rex_relational_similarity/test.jsonl", "w") as f:
|
40 |
-
# f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
41 |
-
#
|
42 |
-
# data = load_dataset("relbert/t_rex_relational_similarity", "filter_unified.min_entity_4_max_predicate_100", split="validation")
|
43 |
-
# df = data.to_pandas()
|
44 |
-
# df['negatives'] = [list(chain(
|
45 |
-
# *[[y.tolist() for y in x.tolist()] for x in df[df.relation_type != i]['positives'].tolist()] +
|
46 |
-
# [[y.tolist() for y in x.tolist()] for x in df[df.relation_type == i]['negatives'].tolist()])) for i in
|
47 |
-
# df['relation_type']]
|
48 |
-
# analogy_data = []
|
49 |
-
# for _, i in df.iterrows():
|
50 |
-
# if len(i['positives']) < 5:
|
51 |
-
# continue
|
52 |
-
# for m, (q, c) in enumerate(combinations(i['positives'], 2)):
|
53 |
-
# if m > 5:
|
54 |
-
# break
|
55 |
-
# negative = i['negatives']
|
56 |
-
# for n in range(3):
|
57 |
-
# seed(n)
|
58 |
-
# shuffle(negative)
|
59 |
-
# analogy_data.append({
|
60 |
-
# "stem": q.tolist(), "choice": [c.tolist()] + negative[:5], "answer": 0, "prefix": i["relation_type"]
|
61 |
-
# })
|
62 |
-
# os.makedirs("dataset/t_rex_relational_similarity", exist_ok=True)
|
63 |
-
# with open("dataset/t_rex_relational_similarity/valid.jsonl", "w") as f:
|
64 |
-
# f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
65 |
-
#
|
66 |
-
# # create analogy from `relbert/conceptnet_relational_similarity`
|
67 |
-
# for s in ['test', 'validation']:
|
68 |
-
# data = load_dataset("relbert/conceptnet_relational_similarity", split=s)
|
69 |
-
# df = data.to_pandas()
|
70 |
-
# df['negatives'] = [list(chain(
|
71 |
-
# *[[y.tolist() for y in x.tolist()] for x in df[df.relation_type != i]['positives'].tolist()] +
|
72 |
-
# [[y.tolist() for y in x.tolist()] for x in df[df.relation_type == i]['negatives'].tolist()])) for i in
|
73 |
-
# df['relation_type']]
|
74 |
-
#
|
75 |
-
# analogy_data = []
|
76 |
-
#
|
77 |
-
# for _, i in df.iterrows():
|
78 |
-
#
|
79 |
-
# if len(i['positives']) < 2:
|
80 |
-
# continue
|
81 |
-
# for m, (q, c) in enumerate(combinations(i['positives'], 2)):
|
82 |
-
# if m > 5:
|
83 |
-
# break
|
84 |
-
# negative = i['negatives']
|
85 |
-
# for n in range(6):
|
86 |
-
# seed(n)
|
87 |
-
# shuffle(negative)
|
88 |
-
# analogy_data.append({
|
89 |
-
# "stem": q.tolist(), "choice": [c.tolist()] + negative[:5], "answer": 0, "prefix": i["relation_type"]
|
90 |
-
# })
|
91 |
-
# print(len(analogy_data))
|
92 |
-
# os.makedirs("dataset/conceptnet_relational_similarity", exist_ok=True)
|
93 |
-
# with open(f"dataset/conceptnet_relational_similarity/{s if s == 'test' else 'valid'}.jsonl", "w") as f:
|
94 |
-
# f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
|
|
1 |
import json
|
2 |
import os
|
3 |
+
from itertools import combinations
|
4 |
from random import shuffle, seed
|
5 |
+
|
6 |
+
import pandas as pd
|
7 |
from datasets import load_dataset
|
8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
+
def get_stats(filename):
|
11 |
+
with open(filename) as f:
|
12 |
+
_data = [json.loads(i) for i in f.read().splitlines()]
|
13 |
+
return len(_data), list(set([len(i['choice']) for i in _data])), len(list(set([i['prefix'] for i in _data])))
|
14 |
+
|
15 |
+
|
16 |
+
def lexical_overlap(word_a, word_b):
|
17 |
+
for a in word_a.split(" "):
|
18 |
+
for b in word_b.split(" "):
|
19 |
+
if a.lower() == b.lower():
|
20 |
+
return True
|
21 |
+
return False
|
22 |
+
|
23 |
+
|
24 |
+
def create_analogy(_data, output_path, negative_per_relation, instance_per_relation=100):
|
25 |
+
# if os.path.exists(output_path):
|
26 |
+
# return
|
27 |
+
df = _data.to_pandas()
|
28 |
+
analogy_data = []
|
29 |
+
for _, i in df.iterrows():
|
30 |
+
target = [(q.tolist(), c.tolist()) for q, c in combinations(i['positives'], 2)
|
31 |
+
if not any(lexical_overlap(c[0], y) or lexical_overlap(c[1], y) for y in q)]
|
32 |
+
if len(target) == 0:
|
33 |
+
continue
|
34 |
+
if len(target) > instance_per_relation:
|
35 |
+
seed(42)
|
36 |
+
shuffle(target)
|
37 |
+
target = target[:instance_per_relation]
|
38 |
+
for m, (q, c) in enumerate(target):
|
39 |
+
negative = []
|
40 |
+
for r in df['relation_type']:
|
41 |
+
if r == i['relation_type']:
|
42 |
+
continue
|
43 |
+
target_per_relation = [y.tolist() for y in df[df['relation_type'] == r]['positives'].values[0]]
|
44 |
+
shuffle(target_per_relation)
|
45 |
+
negative += target_per_relation[:negative_per_relation]
|
46 |
+
analogy_data.append({
|
47 |
+
"stem": q,
|
48 |
+
"choice": [c, c[::-1]] + negative,
|
49 |
+
"answer": 0,
|
50 |
+
"prefix": i["relation_type"]
|
51 |
+
})
|
52 |
+
os.makedirs(os.path.dirname(output_path), exist_ok=True)
|
53 |
+
with open(output_path, "w") as f:
|
54 |
+
f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
55 |
+
|
56 |
+
stat = []
|
57 |
+
###################################################################
|
58 |
+
# create analogy from `relbert/semeval2012_relational_similarity` #
|
59 |
+
###################################################################
|
60 |
+
if not os.path.exists("dataset/semeval2012_relational_similarity/valid.jsonl"):
|
61 |
+
data = load_dataset("relbert/semeval2012_relational_similarity", split="validation")
|
62 |
+
analogy_data = [{
|
63 |
+
"stem": i['positives'][0], "choice": i["negatives"] + [i['positives'][1]], "answer": 2, "prefix": i["relation_type"]
|
64 |
+
} for i in data]
|
65 |
+
os.makedirs("dataset/semeval2012_relational_similarity", exist_ok=True)
|
66 |
+
with open("dataset/semeval2012_relational_similarity/valid.jsonl", "w") as f:
|
67 |
+
f.write("\n".join([json.dumps(i) for i in analogy_data]))
|
68 |
+
|
69 |
+
v_size, v_num_choice, v_relation_type = get_stats("dataset/semeval2012_relational_similarity/valid.jsonl")
|
70 |
+
stat.append({
|
71 |
+
"name": "`semeval2012_relational_similarity`",
|
72 |
+
"Size (valid/test)": f"{v_size}/-",
|
73 |
+
"Num of choice (valid/test)": f"{','.join([str(n) for n in v_num_choice])}/-",
|
74 |
+
"Num of relation group (valid/test)": f"{v_relation_type}/-",
|
75 |
+
"Original Reference": "[relbert/semeval2012_relational_similarity](https://huggingface.co/datasets/relbert/semeval2012_relational_similarity)"
|
76 |
+
})
|
77 |
+
|
78 |
+
|
79 |
+
#############################################################
|
80 |
+
# create analogy from `relbert/t_rex_relational_similarity` #
|
81 |
+
#############################################################
|
82 |
+
data = load_dataset("relbert/t_rex_relational_similarity", "filter_unified.min_entity_1_max_predicate_100", split="test")
|
83 |
+
create_analogy(data, "dataset/t_rex_relational_similarity/test.jsonl", negative_per_relation=2)
|
84 |
+
data = load_dataset("relbert/t_rex_relational_similarity", "filter_unified.min_entity_4_max_predicate_100", split="validation")
|
85 |
+
create_analogy(data, "dataset/t_rex_relational_similarity/valid.jsonl", negative_per_relation=1)
|
86 |
+
|
87 |
+
t_size, t_num_choice, t_relation_type = get_stats("dataset/t_rex_relational_similarity/test.jsonl")
|
88 |
+
v_size, v_num_choice, v_relation_type = get_stats("dataset/t_rex_relational_similarity/valid.jsonl")
|
89 |
+
stat.append({
|
90 |
+
"name": "`t_rex_relational_similarity`",
|
91 |
+
"Size (valid/test)": f"{v_size}/{t_size}",
|
92 |
+
"Num of choice (valid/test)": f"{','.join([str(n) for n in v_num_choice])}/{','.join([str(n) for n in t_num_choice])}",
|
93 |
+
"Num of relation group (valid/test)": f"{v_relation_type}/{t_relation_type}",
|
94 |
+
"Original Reference": "[relbert/t_rex_relational_similarity](https://huggingface.co/datasets/relbert/t_rex_relational_similarity)"
|
95 |
+
})
|
96 |
|
97 |
+
##################################################################
|
98 |
+
# create analogy from `relbert/conceptnet_relational_similarity` #
|
99 |
+
##################################################################
|
100 |
+
data = load_dataset("relbert/conceptnet_relational_similarity", split="test")
|
101 |
+
create_analogy(data, "dataset/conceptnet_relational_similarity/test.jsonl", negative_per_relation=1)
|
102 |
+
data = load_dataset("relbert/conceptnet_relational_similarity", split="validation")
|
103 |
+
create_analogy(data, "dataset/conceptnet_relational_similarity/valid.jsonl", negative_per_relation=1)
|
104 |
+
t_size, t_num_choice, t_relation_type = get_stats("dataset/conceptnet_relational_similarity/test.jsonl")
|
105 |
+
v_size, v_num_choice, v_relation_type = get_stats("dataset/conceptnet_relational_similarity/valid.jsonl")
|
106 |
+
stat.append({
|
107 |
+
"name": "`conceptnet_relational_similarity`",
|
108 |
+
"Size (valid/test)": f"{v_size}/{t_size}",
|
109 |
+
"Num of choice (valid/test)": f"{','.join([str(n) for n in v_num_choice])}/{','.join([str(n) for n in t_num_choice])}",
|
110 |
+
"Num of relation group (valid/test)": f"{v_relation_type}/{t_relation_type}",
|
111 |
+
"Original Reference": "[relbert/conceptnet_relational_similarity](https://huggingface.co/datasets/relbert/conceptnet_relational_similarity)"
|
112 |
+
})
|
113 |
+
print(pd.DataFrame(stat).to_markdown(index=False))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dataset/conceptnet_relational_similarity/test.jsonl
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
dataset/conceptnet_relational_similarity/valid.jsonl
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
dataset/t_rex_relational_similarity/test.jsonl
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
dataset/t_rex_relational_similarity/valid.jsonl
CHANGED
The diff for this file is too large to render.
See raw diff
|
|