File size: 10,595 Bytes
28c8270 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 |
import random
from pathlib import Path
import datasets
from datasets import Value, Sequence, ClassLabel, Features
_CITATION = """\
coming soon
"""
_DESCRIPTION = """\
German Legal Sentences (GLS) is an automatically generated training dataset for semantic sentence
matching in the domain in german legal documents. It follows the concept of weak supervision, where
imperfect labels are generated using multiple heuristics. For this purpose we use a combination of
legal citation matching and BM25 similarity. The contained sentences and their citations are parsed
from real judicial decisions provided by [Open Legal Data](http://openlegaldata.io/)
"""
_VERSION = "0.0.2"
_DATA_URL = f"http://lavis.cs.hs-rm.de/storage/german-legal-sentences/GermanLegalSentences_v{_VERSION}.zip"
class GLSConfig(datasets.BuilderConfig):
"""BuilderConfig."""
def __init__(
self,
load_collection,
load_es_neighbors=None,
n_es_neighbors=None,
**kwargs,
):
"""BuilderConfig.
Args:
**kwargs: keyword arguments forwarded to super.
"""
super(GLSConfig, self).__init__(**kwargs)
self.load_collection = load_collection
self.load_es_neighbors = load_es_neighbors
self.n_es_neighbors = n_es_neighbors
class GermanLegalSentences(datasets.GeneratorBasedBuilder):
BUILDER_CONFIGS = [
GLSConfig(
name="sentences",
load_es_neighbors=False,
load_collection=False,
version=datasets.Version(_VERSION, ""),
description="Just the sentences and their masked references",
),
GLSConfig(
name="pairs",
load_es_neighbors=False,
load_collection=True,
version=datasets.Version(_VERSION, ""),
description="Sentence pairs sharing references",
),
GLSConfig(
name="pairs+es",
load_es_neighbors=True,
load_collection=True,
n_es_neighbors=5,
version=datasets.Version(_VERSION, ""),
description="Sentence pairs sharing references plus ES neighbors",
),
]
def _features(self):
if self.config.name == "sentences":
return datasets.Features(
{
"sent_id": Value("uint32"),
"doc_id": Value("uint32"),
"text": Value("string"),
"references": Sequence(
{
"ref_id": Value("uint32"),
"name": Value("string"),
"type": ClassLabel(names=["AZ", "LAW"]),
}
),
}
)
elif self.config.name == "pairs":
return Features(
{
"query.sent_id": Value("uint32"),
"query.doc_id": Value("uint32"),
"query.text": Value("string"),
"query.ref_ids": Sequence(Value("uint32")),
"related.sent_id": Value("uint32"),
"related.doc_id": Value("uint32"),
"related.text": Value("string"),
"related.ref_ids": Sequence(Value("uint32")),
}
)
elif self.config.name == "pairs+es":
return Features(
{
"query.sent_id": Value("uint32"),
"query.doc_id": Value("uint32"),
"query.text": Value("string"),
"query.ref_ids": Sequence(Value("uint32")),
"related.sent_id": Value("uint32"),
"related.doc_id": Value("uint32"),
"related.text": Value("string"),
"related.ref_ids": Sequence(Value("uint32")),
"es_neighbors.text": Sequence(Value("string")),
"es_neighbors.sent_id": Sequence(Value("uint32")),
"es_neighbors.doc_id": Sequence(Value("uint32")),
"es_neighbors.ref_ids": Sequence(
Sequence(datasets.Value("uint32"))
),
}
)
assert True
def _info(self):
return datasets.DatasetInfo(
description=_DESCRIPTION,
features=self._features(),
supervised_keys=None,
homepage="",
citation=_CITATION,
)
def _split_generators(self, dl_manager):
if dl_manager.manual_dir:
data_dir = Path(dl_manager.manual_dir)
else:
data_dir = Path(dl_manager.download_and_extract(_DATA_URL))
collection = _load_collection(data_dir) if self.config.load_collection else None
sent_ref_map = _load_sent_references(data_dir)
references = (
_load_reference_info(data_dir) if self.config.name == "sentences" else None
)
es_neighbors = (
_load_es_neighbors(data_dir) if self.config.load_es_neighbors else None
)
gen_kwargs = dict()
for split in ("train", "valid", "test"):
gen_kwargs[split] = {
"collection": collection,
"pair_id_file": data_dir / f"{split}.pairs.tsv",
"sentence_file": data_dir / f"{split}.sentences.tsv",
"references": references,
"sent_ref_map": sent_ref_map,
"es_neighbors": es_neighbors,
}
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN, gen_kwargs=gen_kwargs["train"]
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION, gen_kwargs=gen_kwargs["valid"]
),
datasets.SplitGenerator(
name=datasets.Split.TEST, gen_kwargs=gen_kwargs["test"]
),
]
def _generate_examples(self, **kwargs):
if self.config.name.startswith("pairs"):
yield from self._generate_pairs(**kwargs)
elif self.config.name == "sentences":
yield from self._generate_sentences(**kwargs)
else:
assert True
def _generate_pairs(
self, pair_id_file, collection, sent_ref_map, es_neighbors, **kwargs
):
random.seed(17)
with open(pair_id_file, encoding="utf-8") as r:
idx = 0
for line in r:
stripped = line.rstrip()
if stripped:
a, b = stripped.split("\t")
features = {
"query.sent_id": int(a),
"query.doc_id": int(collection[a]["doc_id"]),
"query.text": collection[a]["text"],
"query.ref_ids": sent_ref_map[a],
"related.sent_id": int(b),
"related.doc_id": int(collection[b]["doc_id"]),
"related.text": collection[b]["text"],
"related.ref_ids": sent_ref_map[b],
}
if self.config.name == "pairs+es":
curr_es_neighbors = es_neighbors.get(a) or []
if len(curr_es_neighbors) < self.config.n_es_neighbors:
continue
es_sent_ids = random.sample(
curr_es_neighbors, k=self.config.n_es_neighbors
)
additional_features = {
"es_neighbors.sent_id": [int(i) for i in es_sent_ids],
"es_neighbors.doc_id": [
int(collection[i]["doc_id"]) for i in es_sent_ids
],
"es_neighbors.text": [
collection[i]["text"] for i in es_sent_ids
],
"es_neighbors.ref_ids": [
sent_ref_map[i] for i in es_sent_ids
],
}
features.update(additional_features)
yield idx, features
idx += 1
def _generate_sentences(
self,
sentence_file,
references,
sent_ref_map,
**kwargs,
):
with open(sentence_file, encoding="utf-8") as r:
for idx, line in enumerate(r):
stripped = line.rstrip()
if stripped == "":
continue
s_id, doc_id, text = stripped.split("\t", maxsplit=2)
yield idx, {
"sent_id": int(s_id),
"doc_id": int(doc_id),
"text": text,
"references": [
{
"ref_id": int(r_id),
"name": references[r_id][1],
"type": references[r_id][0],
}
for r_id in sent_ref_map[s_id]
],
}
def _load_collection(data_dir):
collection = dict()
for split in ("train", "valid", "test"):
with open(data_dir / f"{split}.sentences.tsv", encoding="utf-8") as r:
for line in r:
s_id, d_id, sent = line.strip().split("\t", maxsplit=2)
collection[s_id] = {"doc_id": d_id, "text": sent}
return collection
def _load_reference_info(data_dir):
with open(data_dir / "refs.tsv", encoding="utf-8") as r:
references = {
r_id: (r_type, r_name.rstrip())
for r_id, r_type, r_name in (
line.split("\t", maxsplit=2) for line in r if len(line) > 2
)
}
return references
def _load_sent_references(data_dir):
with open(data_dir / "sent_ref_map.tsv", encoding="utf-8") as r:
sent_ref_map = {
s_id: r_ids.rstrip().split()
for s_id, r_ids in (
line.split("\t", maxsplit=1) for line in r if len(line) > 2
)
}
return sent_ref_map
def _load_es_neighbors(data_dir):
with open(data_dir / "es_neighbors.tsv", encoding="utf-8") as r:
es_neighbors = {
s_id: other_s_ids.rstrip().split()
for s_id, other_s_ids in (
line.split("\t", maxsplit=1) for line in r if len(line) > 2
)
}
return es_neighbors
|