add extraction script
Browse files- extract_data.py +94 -0
extract_data.py
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Script to generate splits for benchmarking text embedding clustering.
|
2 |
+
Based on data from GermEval 2019 Shared Task on Hierarchical Tesk Classification (https://www.inf.uni-hamburg.de/en/inst/ab/lt/resources/data/germeval-2019-hmc.html)."""
|
3 |
+
|
4 |
+
import os
|
5 |
+
import random
|
6 |
+
import sys
|
7 |
+
from collections import Counter
|
8 |
+
|
9 |
+
import jsonlines
|
10 |
+
import numpy as np
|
11 |
+
import pandas as pd
|
12 |
+
from bs4 import BeautifulSoup
|
13 |
+
|
14 |
+
random.seed(42)
|
15 |
+
|
16 |
+
# path to "data" folder, can be retrieved from here: https://www.inf.uni-hamburg.de/en/inst/ab/lt/resources/data/germeval-2019-hmc/germeval2019t1-public-data-final.zip
|
17 |
+
DATA_PATH = sys.argv[1]
|
18 |
+
|
19 |
+
INCLUDE_BODY = (
|
20 |
+
True # True: combine title and article body (p2p), False: only title (s2s)
|
21 |
+
)
|
22 |
+
|
23 |
+
NUM_SPLITS = 10
|
24 |
+
SPLIT_RANGE = np.array([0.1, 1.0])
|
25 |
+
|
26 |
+
|
27 |
+
def get_samples(soup, include_body=INCLUDE_BODY):
|
28 |
+
d1_counter = Counter([d1.string for d1 in soup.find_all("topic", {"d": 1})])
|
29 |
+
|
30 |
+
samples = []
|
31 |
+
for book in soup.find_all("book"):
|
32 |
+
if book.title.string is None or book.body.string is None:
|
33 |
+
continue
|
34 |
+
|
35 |
+
d0_topics = list(set([d.string for d in book.find_all("topic", {"d": 0})]))
|
36 |
+
d1_topics = list(set([d.string for d in book.find_all("topic", {"d": 1})]))
|
37 |
+
|
38 |
+
if len(d0_topics) != 1:
|
39 |
+
continue
|
40 |
+
if len(d1_topics) < 1 or len(d1_topics) > 2:
|
41 |
+
continue
|
42 |
+
|
43 |
+
d0_label = d0_topics[0]
|
44 |
+
d1_label = sorted(d1_topics, key=lambda x: d1_counter[x])[0]
|
45 |
+
|
46 |
+
text = book.title.string
|
47 |
+
if include_body:
|
48 |
+
text += "\n" + book.body.string
|
49 |
+
|
50 |
+
samples.append([text, d0_label, d1_label])
|
51 |
+
|
52 |
+
return pd.DataFrame(samples, columns=["sentences", "d0_label", "d1_label"])
|
53 |
+
|
54 |
+
|
55 |
+
def get_split(frame, label="d0_label", split_range=SPLIT_RANGE):
|
56 |
+
samples = random.randint(*(split_range * len(frame)).astype(int))
|
57 |
+
return (
|
58 |
+
frame.sample(samples)[["sentences", label]]
|
59 |
+
.rename(columns={label: "labels"})[["sentences", "labels"]]
|
60 |
+
.to_dict("list")
|
61 |
+
)
|
62 |
+
|
63 |
+
|
64 |
+
def write_sets(name, sets):
|
65 |
+
with jsonlines.open(name, "w") as f_out:
|
66 |
+
f_out.write_all(sets)
|
67 |
+
|
68 |
+
|
69 |
+
train = open(os.path.join(DATA_PATH, "blurbs_train.txt"), encoding="utf-8").read()
|
70 |
+
dev = open(os.path.join(DATA_PATH, "blurbs_dev.txt"), encoding="utf-8").read()
|
71 |
+
test = open(os.path.join(DATA_PATH, "blurbs_test.txt"), encoding="utf-8").read()
|
72 |
+
|
73 |
+
soup = BeautifulSoup(train + "\n\n" + dev + "\n\n" + test, "html.parser")
|
74 |
+
|
75 |
+
samples = get_samples(soup)
|
76 |
+
|
77 |
+
sets = []
|
78 |
+
# coarse clustering
|
79 |
+
for _ in range(NUM_SPLITS):
|
80 |
+
sets.append(get_split(samples))
|
81 |
+
|
82 |
+
# fine grained clustering inside top-level category (d0)
|
83 |
+
for d0 in samples["d0_label"].unique():
|
84 |
+
sets.append(
|
85 |
+
(samples[samples.d0_label == d0])
|
86 |
+
.rename(columns={"d1_label": "labels"})[["sentences", "labels"]]
|
87 |
+
.to_dict("list")
|
88 |
+
)
|
89 |
+
|
90 |
+
# fine grained clustering
|
91 |
+
for _ in range(NUM_SPLITS):
|
92 |
+
sets.append(get_split(samples, label="d1_label"))
|
93 |
+
|
94 |
+
write_sets("test.jsonl", sets)
|