fix dataset
Browse files- README.md +2 -4
- process/tweet_ner.py +62 -0
README.md
CHANGED
@@ -62,14 +62,12 @@ The data fields are the same among all splits.
|
|
62 |
#### tweet_qa
|
63 |
- `text`: a `string` feature.
|
64 |
- `gold_label_str`: a `string` feature.
|
65 |
-
- `
|
66 |
-
- `question`: a `string` feature.
|
67 |
|
68 |
#### tweet_qg
|
69 |
- `text`: a `string` feature.
|
70 |
- `gold_label_str`: a `string` feature.
|
71 |
-
- `
|
72 |
-
- `question`: a `string` feature.
|
73 |
|
74 |
#### tweet_intimacy
|
75 |
- `text`: a `string` feature.
|
|
|
62 |
#### tweet_qa
|
63 |
- `text`: a `string` feature.
|
64 |
- `gold_label_str`: a `string` feature.
|
65 |
+
- `context`: a `string` feature.
|
|
|
66 |
|
67 |
#### tweet_qg
|
68 |
- `text`: a `string` feature.
|
69 |
- `gold_label_str`: a `string` feature.
|
70 |
+
- `context`: a `string` feature.
|
|
|
71 |
|
72 |
#### tweet_intimacy
|
73 |
- `text`: a `string` feature.
|
process/tweet_ner.py
CHANGED
@@ -1,7 +1,66 @@
|
|
1 |
import os
|
2 |
import json
|
|
|
|
|
3 |
from datasets import load_dataset
|
4 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
os.makedirs("data/tweet_ner7", exist_ok=True)
|
6 |
data = load_dataset("tner/tweetner7")
|
7 |
|
@@ -10,6 +69,9 @@ def process(tmp):
|
|
10 |
tmp = [i.to_dict() for _, i in tmp.iterrows()]
|
11 |
for i in tmp:
|
12 |
i.pop("id")
|
|
|
|
|
|
|
13 |
i['gold_label_sequence'] = i.pop('tags').tolist()
|
14 |
i['text_tokenized'] = i.pop('tokens').tolist()
|
15 |
i['text'] = ' '.join(i['text_tokenized'])
|
|
|
1 |
import os
|
2 |
import json
|
3 |
+
from typing import List
|
4 |
+
from pprint import pprint
|
5 |
from datasets import load_dataset
|
6 |
|
7 |
+
label2id = {
|
8 |
+
"B-corporation": 0,
|
9 |
+
"B-creative_work": 1,
|
10 |
+
"B-event": 2,
|
11 |
+
"B-group": 3,
|
12 |
+
"B-location": 4,
|
13 |
+
"B-person": 5,
|
14 |
+
"B-product": 6,
|
15 |
+
"I-corporation": 7,
|
16 |
+
"I-creative_work": 8,
|
17 |
+
"I-event": 9,
|
18 |
+
"I-group": 10,
|
19 |
+
"I-location": 11,
|
20 |
+
"I-person": 12,
|
21 |
+
"I-product": 13,
|
22 |
+
"O": 14
|
23 |
+
}
|
24 |
+
|
25 |
+
|
26 |
+
def decode_ner_tags(tag_sequence: List, input_sequence: List):
|
27 |
+
""" decode ner tag sequence """
|
28 |
+
def update_collection(_tmp_entity, _tmp_entity_type, _tmp_pos, _out):
|
29 |
+
if len(_tmp_entity) != 0 and _tmp_entity_type is not None:
|
30 |
+
_out.append({'type': _tmp_entity_type, 'entity': _tmp_entity, 'position': _tmp_pos})
|
31 |
+
_tmp_entity = []
|
32 |
+
_tmp_entity_type = None
|
33 |
+
return _tmp_entity, _tmp_entity_type, _tmp_pos, _out
|
34 |
+
|
35 |
+
assert len(tag_sequence) == len(input_sequence), str([len(tag_sequence), len(input_sequence)])
|
36 |
+
out = []
|
37 |
+
tmp_entity = []
|
38 |
+
tmp_pos = []
|
39 |
+
tmp_entity_type = None
|
40 |
+
for n, (_l, _i) in enumerate(zip(tag_sequence, input_sequence)):
|
41 |
+
if _l.startswith('B-'):
|
42 |
+
_, _, _, out = update_collection(tmp_entity, tmp_entity_type, tmp_pos, out)
|
43 |
+
tmp_entity_type = '-'.join(_l.split('-')[1:])
|
44 |
+
tmp_entity = [_i]
|
45 |
+
tmp_pos = [n]
|
46 |
+
elif _l.startswith('I-'):
|
47 |
+
tmp_tmp_entity_type = '-'.join(_l.split('-')[1:])
|
48 |
+
if len(tmp_entity) == 0:
|
49 |
+
# if 'I' not start with 'B', skip it
|
50 |
+
tmp_entity, tmp_entity_type, tmp_pos, out = update_collection(tmp_entity, tmp_entity_type, tmp_pos, out)
|
51 |
+
elif tmp_tmp_entity_type != tmp_entity_type:
|
52 |
+
# if the type does not match with the B, skip
|
53 |
+
tmp_entity, tmp_entity_type, tmp_pos, out = update_collection(tmp_entity, tmp_entity_type, tmp_pos, out)
|
54 |
+
else:
|
55 |
+
tmp_entity.append(_i)
|
56 |
+
tmp_pos.append(n)
|
57 |
+
elif _l == 'O':
|
58 |
+
tmp_entity, tmp_entity_type, tmp_pos, out = update_collection(tmp_entity, tmp_entity_type, tmp_pos, out)
|
59 |
+
else:
|
60 |
+
raise ValueError('unknown tag: {}'.format(_l))
|
61 |
+
_, _, _, out = update_collection(tmp_entity, tmp_entity_type, tmp_pos, out)
|
62 |
+
return out
|
63 |
+
|
64 |
os.makedirs("data/tweet_ner7", exist_ok=True)
|
65 |
data = load_dataset("tner/tweetner7")
|
66 |
|
|
|
69 |
tmp = [i.to_dict() for _, i in tmp.iterrows()]
|
70 |
for i in tmp:
|
71 |
i.pop("id")
|
72 |
+
entities = decode_ner_tags(i['tags'].tolist(), i['tokens'].tolist())
|
73 |
+
pprint(entities)
|
74 |
+
input()
|
75 |
i['gold_label_sequence'] = i.pop('tags').tolist()
|
76 |
i['text_tokenized'] = i.pop('tokens').tolist()
|
77 |
i['text'] = ' '.join(i['text_tokenized'])
|