Create paperswithcode-aspects.py
Browse files- paperswithcode-aspects.py +156 -0
paperswithcode-aspects.py
ADDED
@@ -0,0 +1,156 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import absolute_import, division, print_function
|
2 |
+
|
3 |
+
import json
|
4 |
+
import os
|
5 |
+
import sys
|
6 |
+
|
7 |
+
import datasets
|
8 |
+
from pyarrow import csv
|
9 |
+
|
10 |
+
_DESCRIPTION = """Papers with aspects from paperswithcode.com dataset"""
|
11 |
+
|
12 |
+
_HOMEPAGE = "https://github.com/malteos/aspect-document-embeddings"
|
13 |
+
|
14 |
+
_CITATION = '''@InProceedings{Ostendorff2022,
|
15 |
+
title = {Specialized Document Embeddings for Aspect-based Similarity of Research Papers},
|
16 |
+
booktitle = {Proceedings of the {ACM}/{IEEE} {Joint} {Conference} on {Digital} {Libraries} ({JCDL})},
|
17 |
+
author = {Ostendorff, Malte and Blume, Till, Ruas, Terry and Gipp, Bela and Rehm, Georg},
|
18 |
+
year = {2022},
|
19 |
+
}'''
|
20 |
+
|
21 |
+
DATA_URL = "http://datasets.fiq.de/paperswithcode_aspects.tar.gz"
|
22 |
+
|
23 |
+
DOC_A_COL = "from_paper_id"
|
24 |
+
DOC_B_COL = "to_paper_id"
|
25 |
+
LABEL_COL = "label"
|
26 |
+
|
27 |
+
# binary classification (y=similar, n=dissimilar)
|
28 |
+
LABEL_CLASSES = labels = ['y', 'n']
|
29 |
+
|
30 |
+
ASPECTS = ['task', 'method', 'dataset']
|
31 |
+
|
32 |
+
|
33 |
+
def get_train_split(aspect, k):
|
34 |
+
return datasets.Split(f'fold_{aspect}_{k}_train')
|
35 |
+
|
36 |
+
|
37 |
+
def get_test_split(aspect, k):
|
38 |
+
return datasets.Split(f'fold_{aspect}_{k}_test')
|
39 |
+
|
40 |
+
|
41 |
+
class PWCConfig(datasets.BuilderConfig):
|
42 |
+
def __init__(self, features, data_url, aspects, **kwargs):
|
43 |
+
super().__init__(version=datasets.Version("0.1.0"), **kwargs)
|
44 |
+
self.features = features
|
45 |
+
self.data_url = data_url
|
46 |
+
self.aspects = aspects
|
47 |
+
|
48 |
+
|
49 |
+
class PWCAspects(datasets.GeneratorBasedBuilder):
|
50 |
+
"""Paper aspects dataset."""
|
51 |
+
|
52 |
+
BUILDER_CONFIGS = [
|
53 |
+
PWCConfig(
|
54 |
+
name="docs",
|
55 |
+
description="document text and meta data",
|
56 |
+
# Metadata format from paperswithcode.com
|
57 |
+
# see https://github.com/paperswithcode/paperswithcode-data
|
58 |
+
features={
|
59 |
+
"paper_id": datasets.Value("string"),
|
60 |
+
"paper_url": datasets.Value("string"),
|
61 |
+
"title": datasets.Value("string"),
|
62 |
+
"abstract": datasets.Value("string"),
|
63 |
+
"arxiv_id": datasets.Value("string"),
|
64 |
+
"url_abs": datasets.Value("string"),
|
65 |
+
"url_pdf": datasets.Value("string"),
|
66 |
+
"aspect_tasks": datasets.Sequence(datasets.Value('string', id='task')),
|
67 |
+
"aspect_methods": datasets.Sequence(datasets.Value('string', id='method')),
|
68 |
+
"aspect_datasets": datasets.Sequence(datasets.Value('string', id='dataset')),
|
69 |
+
},
|
70 |
+
data_url=DATA_URL,
|
71 |
+
aspects=ASPECTS,
|
72 |
+
),
|
73 |
+
PWCConfig(
|
74 |
+
name="relations",
|
75 |
+
description=" relation data",
|
76 |
+
features={
|
77 |
+
DOC_A_COL: datasets.Value("string"),
|
78 |
+
DOC_B_COL: datasets.Value("string"),
|
79 |
+
LABEL_COL: datasets.Value("string"),
|
80 |
+
},
|
81 |
+
data_url=DATA_URL,
|
82 |
+
aspects=ASPECTS,
|
83 |
+
),
|
84 |
+
]
|
85 |
+
|
86 |
+
def _info(self):
|
87 |
+
return datasets.DatasetInfo(
|
88 |
+
description=_DESCRIPTION + self.config.description,
|
89 |
+
features=datasets.Features(self.config.features),
|
90 |
+
homepage=_HOMEPAGE,
|
91 |
+
citation=_CITATION,
|
92 |
+
)
|
93 |
+
|
94 |
+
def _split_generators(self, dl_manager):
|
95 |
+
arch_path = dl_manager.download_and_extract(self.config.data_url)
|
96 |
+
|
97 |
+
if "relations" in self.config.name:
|
98 |
+
train_file = "train.csv"
|
99 |
+
test_file = "test.csv"
|
100 |
+
|
101 |
+
generators = []
|
102 |
+
|
103 |
+
# for k in [1, 2, 3, 4]:
|
104 |
+
for aspect in self.config.aspects:
|
105 |
+
for k in ["sample"] + [1, 2, 3, 4]:
|
106 |
+
folds_path = os.path.join(arch_path, 'folds', aspect, str(k))
|
107 |
+
generators += [
|
108 |
+
datasets.SplitGenerator(
|
109 |
+
name=get_train_split(aspect, k),
|
110 |
+
gen_kwargs={'filepath': os.path.join(folds_path, train_file)}
|
111 |
+
),
|
112 |
+
datasets.SplitGenerator(
|
113 |
+
name=get_test_split(aspect, k),
|
114 |
+
gen_kwargs={'filepath': os.path.join(folds_path, test_file)}
|
115 |
+
)
|
116 |
+
]
|
117 |
+
return generators
|
118 |
+
|
119 |
+
elif "docs" in self.config.name:
|
120 |
+
# docs
|
121 |
+
docs_file = os.path.join(arch_path, "docs.jsonl")
|
122 |
+
|
123 |
+
return [
|
124 |
+
datasets.SplitGenerator(name=datasets.Split('docs'), gen_kwargs={"filepath": docs_file}),
|
125 |
+
]
|
126 |
+
else:
|
127 |
+
raise ValueError()
|
128 |
+
|
129 |
+
@staticmethod
|
130 |
+
def get_dict_value(d, key, default=None):
|
131 |
+
if key in d:
|
132 |
+
return d[key]
|
133 |
+
else:
|
134 |
+
return default
|
135 |
+
|
136 |
+
def _generate_examples(self, filepath):
|
137 |
+
"""Generate docs + rel examples."""
|
138 |
+
|
139 |
+
if "relations" in self.config.name:
|
140 |
+
df = csv.read_csv(filepath).to_pandas()
|
141 |
+
|
142 |
+
for idx, row in df.iterrows():
|
143 |
+
yield idx, {
|
144 |
+
DOC_A_COL: str(row[DOC_A_COL]),
|
145 |
+
DOC_B_COL: str(row[DOC_B_COL]),
|
146 |
+
LABEL_COL: row['label'], # !!! labels != label
|
147 |
+
}
|
148 |
+
|
149 |
+
elif self.config.name == "docs":
|
150 |
+
with open(filepath, 'r') as f:
|
151 |
+
for i, line in enumerate(f):
|
152 |
+
doc = json.loads(line)
|
153 |
+
# extract feature keys from doc
|
154 |
+
features = {k: doc[k] if k in doc else None for k in self.config.features.keys()}
|
155 |
+
|
156 |
+
yield i, features
|