wrzlk commited on
Commit
28c8270
1 Parent(s): 2f9c831

Create german_legal_sentences.py

Browse files
Files changed (1) hide show
  1. german_legal_sentences.py +285 -0
german_legal_sentences.py ADDED
@@ -0,0 +1,285 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import random
2
+
3
+ from pathlib import Path
4
+ import datasets
5
+ from datasets import Value, Sequence, ClassLabel, Features
6
+
7
+ _CITATION = """\
8
+ coming soon
9
+ """
10
+
11
+ _DESCRIPTION = """\
12
+ German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence
13
+ matching in the domain in german legal documents. It follows the concept of weak supervision, where
14
+ imperfect labels are generated using multiple heuristics. For this purpose we use a combination of
15
+ legal citation matching and BM25 similarity. The contained sentences and their citations are parsed
16
+ from real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)
17
+ """
18
+
19
+ _VERSION = "0.0.2"
20
+ _DATA_URL = f"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v{_VERSION}.zip"
21
+
22
+
23
+ class GLSConfig(datasets.BuilderConfig):
24
+ """BuilderConfig."""
25
+
26
+ def __init__(
27
+ self,
28
+ load_collection,
29
+ load_es_neighbors=None,
30
+ n_es_neighbors=None,
31
+ **kwargs,
32
+ ):
33
+ """BuilderConfig.
34
+ Args:
35
+ **kwargs: keyword arguments forwarded to super.
36
+ """
37
+ super(GLSConfig, self).__init__(**kwargs)
38
+ self.load_collection = load_collection
39
+ self.load_es_neighbors = load_es_neighbors
40
+ self.n_es_neighbors = n_es_neighbors
41
+
42
+
43
+ class GermanLegalSentences(datasets.GeneratorBasedBuilder):
44
+ BUILDER_CONFIGS = [
45
+ GLSConfig(
46
+ name="sentences",
47
+ load_es_neighbors=False,
48
+ load_collection=False,
49
+ version=datasets.Version(_VERSION, ""),
50
+ description="Just the sentences and their masked references",
51
+ ),
52
+ GLSConfig(
53
+ name="pairs",
54
+ load_es_neighbors=False,
55
+ load_collection=True,
56
+ version=datasets.Version(_VERSION, ""),
57
+ description="Sentence pairs sharing references",
58
+ ),
59
+ GLSConfig(
60
+ name="pairs+es",
61
+ load_es_neighbors=True,
62
+ load_collection=True,
63
+ n_es_neighbors=5,
64
+ version=datasets.Version(_VERSION, ""),
65
+ description="Sentence pairs sharing references plus ES neighbors",
66
+ ),
67
+ ]
68
+
69
+ def _features(self):
70
+ if self.config.name == "sentences":
71
+ return datasets.Features(
72
+ {
73
+ "sent_id": Value("uint32"),
74
+ "doc_id": Value("uint32"),
75
+ "text": Value("string"),
76
+ "references": Sequence(
77
+ {
78
+ "ref_id": Value("uint32"),
79
+ "name": Value("string"),
80
+ "type": ClassLabel(names=["AZ", "LAW"]),
81
+ }
82
+ ),
83
+ }
84
+ )
85
+ elif self.config.name == "pairs":
86
+ return Features(
87
+ {
88
+ "query.sent_id": Value("uint32"),
89
+ "query.doc_id": Value("uint32"),
90
+ "query.text": Value("string"),
91
+ "query.ref_ids": Sequence(Value("uint32")),
92
+ "related.sent_id": Value("uint32"),
93
+ "related.doc_id": Value("uint32"),
94
+ "related.text": Value("string"),
95
+ "related.ref_ids": Sequence(Value("uint32")),
96
+ }
97
+ )
98
+ elif self.config.name == "pairs+es":
99
+ return Features(
100
+ {
101
+ "query.sent_id": Value("uint32"),
102
+ "query.doc_id": Value("uint32"),
103
+ "query.text": Value("string"),
104
+ "query.ref_ids": Sequence(Value("uint32")),
105
+ "related.sent_id": Value("uint32"),
106
+ "related.doc_id": Value("uint32"),
107
+ "related.text": Value("string"),
108
+ "related.ref_ids": Sequence(Value("uint32")),
109
+ "es_neighbors.text": Sequence(Value("string")),
110
+ "es_neighbors.sent_id": Sequence(Value("uint32")),
111
+ "es_neighbors.doc_id": Sequence(Value("uint32")),
112
+ "es_neighbors.ref_ids": Sequence(
113
+ Sequence(datasets.Value("uint32"))
114
+ ),
115
+ }
116
+ )
117
+ assert True
118
+
119
+ def _info(self):
120
+ return datasets.DatasetInfo(
121
+ description=_DESCRIPTION,
122
+ features=self._features(),
123
+ supervised_keys=None,
124
+ homepage="",
125
+ citation=_CITATION,
126
+ )
127
+
128
+ def _split_generators(self, dl_manager):
129
+ if dl_manager.manual_dir:
130
+ data_dir = Path(dl_manager.manual_dir)
131
+ else:
132
+ data_dir = Path(dl_manager.download_and_extract(_DATA_URL))
133
+ collection = _load_collection(data_dir) if self.config.load_collection else None
134
+ sent_ref_map = _load_sent_references(data_dir)
135
+ references = (
136
+ _load_reference_info(data_dir) if self.config.name == "sentences" else None
137
+ )
138
+ es_neighbors = (
139
+ _load_es_neighbors(data_dir) if self.config.load_es_neighbors else None
140
+ )
141
+
142
+ gen_kwargs = dict()
143
+ for split in ("train", "valid", "test"):
144
+ gen_kwargs[split] = {
145
+ "collection": collection,
146
+ "pair_id_file": data_dir / f"{split}.pairs.tsv",
147
+ "sentence_file": data_dir / f"{split}.sentences.tsv",
148
+ "references": references,
149
+ "sent_ref_map": sent_ref_map,
150
+ "es_neighbors": es_neighbors,
151
+ }
152
+ return [
153
+ datasets.SplitGenerator(
154
+ name=datasets.Split.TRAIN, gen_kwargs=gen_kwargs["train"]
155
+ ),
156
+ datasets.SplitGenerator(
157
+ name=datasets.Split.VALIDATION, gen_kwargs=gen_kwargs["valid"]
158
+ ),
159
+ datasets.SplitGenerator(
160
+ name=datasets.Split.TEST, gen_kwargs=gen_kwargs["test"]
161
+ ),
162
+ ]
163
+
164
+ def _generate_examples(self, **kwargs):
165
+ if self.config.name.startswith("pairs"):
166
+ yield from self._generate_pairs(**kwargs)
167
+ elif self.config.name == "sentences":
168
+ yield from self._generate_sentences(**kwargs)
169
+ else:
170
+ assert True
171
+
172
+ def _generate_pairs(
173
+ self, pair_id_file, collection, sent_ref_map, es_neighbors, **kwargs
174
+ ):
175
+ random.seed(17)
176
+ with open(pair_id_file, encoding="utf-8") as r:
177
+ idx = 0
178
+ for line in r:
179
+ stripped = line.rstrip()
180
+ if stripped:
181
+ a, b = stripped.split("\t")
182
+ features = {
183
+ "query.sent_id": int(a),
184
+ "query.doc_id": int(collection[a]["doc_id"]),
185
+ "query.text": collection[a]["text"],
186
+ "query.ref_ids": sent_ref_map[a],
187
+ "related.sent_id": int(b),
188
+ "related.doc_id": int(collection[b]["doc_id"]),
189
+ "related.text": collection[b]["text"],
190
+ "related.ref_ids": sent_ref_map[b],
191
+ }
192
+ if self.config.name == "pairs+es":
193
+ curr_es_neighbors = es_neighbors.get(a) or []
194
+ if len(curr_es_neighbors) < self.config.n_es_neighbors:
195
+ continue
196
+
197
+ es_sent_ids = random.sample(
198
+ curr_es_neighbors, k=self.config.n_es_neighbors
199
+ )
200
+ additional_features = {
201
+ "es_neighbors.sent_id": [int(i) for i in es_sent_ids],
202
+ "es_neighbors.doc_id": [
203
+ int(collection[i]["doc_id"]) for i in es_sent_ids
204
+ ],
205
+ "es_neighbors.text": [
206
+ collection[i]["text"] for i in es_sent_ids
207
+ ],
208
+ "es_neighbors.ref_ids": [
209
+ sent_ref_map[i] for i in es_sent_ids
210
+ ],
211
+ }
212
+ features.update(additional_features)
213
+ yield idx, features
214
+ idx += 1
215
+
216
+ def _generate_sentences(
217
+ self,
218
+ sentence_file,
219
+ references,
220
+ sent_ref_map,
221
+ **kwargs,
222
+ ):
223
+ with open(sentence_file, encoding="utf-8") as r:
224
+ for idx, line in enumerate(r):
225
+ stripped = line.rstrip()
226
+ if stripped == "":
227
+ continue
228
+ s_id, doc_id, text = stripped.split("\t", maxsplit=2)
229
+ yield idx, {
230
+ "sent_id": int(s_id),
231
+ "doc_id": int(doc_id),
232
+ "text": text,
233
+ "references": [
234
+ {
235
+ "ref_id": int(r_id),
236
+ "name": references[r_id][1],
237
+ "type": references[r_id][0],
238
+ }
239
+ for r_id in sent_ref_map[s_id]
240
+ ],
241
+ }
242
+
243
+
244
+ def _load_collection(data_dir):
245
+ collection = dict()
246
+ for split in ("train", "valid", "test"):
247
+ with open(data_dir / f"{split}.sentences.tsv", encoding="utf-8") as r:
248
+ for line in r:
249
+ s_id, d_id, sent = line.strip().split("\t", maxsplit=2)
250
+ collection[s_id] = {"doc_id": d_id, "text": sent}
251
+ return collection
252
+
253
+
254
+ def _load_reference_info(data_dir):
255
+ with open(data_dir / "refs.tsv", encoding="utf-8") as r:
256
+ references = {
257
+ r_id: (r_type, r_name.rstrip())
258
+ for r_id, r_type, r_name in (
259
+ line.split("\t", maxsplit=2) for line in r if len(line) > 2
260
+ )
261
+ }
262
+
263
+ return references
264
+
265
+
266
+ def _load_sent_references(data_dir):
267
+ with open(data_dir / "sent_ref_map.tsv", encoding="utf-8") as r:
268
+ sent_ref_map = {
269
+ s_id: r_ids.rstrip().split()
270
+ for s_id, r_ids in (
271
+ line.split("\t", maxsplit=1) for line in r if len(line) > 2
272
+ )
273
+ }
274
+ return sent_ref_map
275
+
276
+
277
+ def _load_es_neighbors(data_dir):
278
+ with open(data_dir / "es_neighbors.tsv", encoding="utf-8") as r:
279
+ es_neighbors = {
280
+ s_id: other_s_ids.rstrip().split()
281
+ for s_id, other_s_ids in (
282
+ line.split("\t", maxsplit=1) for line in r if len(line) > 2
283
+ )
284
+ }
285
+ return es_neighbors