Commit
•
8ec030d
1
Parent(s):
aa00611
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,158 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- ar
|
4 |
+
- ca
|
5 |
+
- de
|
6 |
+
- en
|
7 |
+
- es
|
8 |
+
- fa
|
9 |
+
- fr
|
10 |
+
- it
|
11 |
+
- ko
|
12 |
+
- nl
|
13 |
+
- pl
|
14 |
+
- pt
|
15 |
+
- ru
|
16 |
+
- sv
|
17 |
+
- uk
|
18 |
+
widget:
|
19 |
+
- text: >-
|
20 |
+
The Red Hot Chili Peppers were formed in Los Angeles by Kiedis, Flea, guitarist Hillel Slovak and drummer Jack Irons.
|
21 |
+
tags:
|
22 |
+
- seq2seq
|
23 |
+
- relation-extraction
|
24 |
+
|
25 |
+
license: cc-by-nc-sa-4.0
|
26 |
+
---
|
27 |
+
# RED^{FM}: a Filtered and Multilingual Relation Extraction Dataset
|
28 |
+
|
29 |
+
This a multilingual version of [REBEL](https://huggingface.co/Babelscape/rebel-large). It can be used as a standalone multulingual Relation Extraction system, or as a pretrained system to be tuned on multilingual Relation Extraction datasets.
|
30 |
+
|
31 |
+
mREBEL is introduced in the ACL 2023 paper [RED^{FM}: a Filtered and Multilingual Relation Extraction Dataset](https://github.com/Babelscape/rebel/blob/main/docs/). We present a new multilingual Relation Extraction dataset and train a multilingual version of REBEL which reframed Relation Extraction as a seq2seq task. The paper can be found [here](https://github.com/Babelscape/rebel/blob/main/docs/). If you use the code or model, please reference this work in your paper:
|
32 |
+
|
33 |
+
@inproceedings{huguet-cabot-et-al-2023-red,
|
34 |
+
title = "RED^{FM}: a Filtered and Multilingual Relation Extraction Dataset",
|
35 |
+
author = "Huguet Cabot, Pere-Llu{\'\i}s and
|
36 |
+
Navigli, Roberto",
|
37 |
+
booktitle = "ACL 2023",
|
38 |
+
month = jul,
|
39 |
+
year = "2023",
|
40 |
+
address = "Toronto, Canada",
|
41 |
+
publisher = "Association for Computational Linguistics",
|
42 |
+
}
|
43 |
+
|
44 |
+
The original repository for the paper can be found [here](https://github.com/Babelscape/rebel)
|
45 |
+
|
46 |
+
Be aware that the inference widget at the right does not output special tokens, which are necessary to distinguish the subject, object and relation types. For a demo of REBEL and its pre-training dataset check the [Spaces demo](https://huggingface.co/spaces/Babelscape/rebel-demo).
|
47 |
+
|
48 |
+
## Pipeline usage
|
49 |
+
|
50 |
+
```python
|
51 |
+
from transformers import pipeline
|
52 |
+
|
53 |
+
triplet_extractor = pipeline('text2text-generation', model='Babelscape/mrebel-large', tokenizer='Babelscape/mrebel-large')
|
54 |
+
# We need to use the tokenizer manually since we need special tokens.
|
55 |
+
extracted_text = triplet_extractor.tokenizer.batch_decode([triplet_extractor("The Red Hot Chili Peppers were formed in Los Angeles by Kiedis, Flea, guitarist Hillel Slovak and drummer Jack Irons.", return_tensors=True, return_text=False)[0]["generated_token_ids"]])
|
56 |
+
print(extracted_text[0])
|
57 |
+
# Function to parse the generated text and extract the triplets
|
58 |
+
def extract_triplets(text):
|
59 |
+
triplets = []
|
60 |
+
relation, subject, relation, object_ = '', '', '', ''
|
61 |
+
text = text.strip()
|
62 |
+
current = 'x'
|
63 |
+
for token in text.replace("<s>", "").replace("<pad>", "").replace("</s>", "").split():
|
64 |
+
if token == "<triplet>":
|
65 |
+
current = 't'
|
66 |
+
if relation != '':
|
67 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
68 |
+
relation = ''
|
69 |
+
subject = ''
|
70 |
+
elif token == "<subj>":
|
71 |
+
current = 's'
|
72 |
+
if relation != '':
|
73 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
74 |
+
object_ = ''
|
75 |
+
elif token == "<obj>":
|
76 |
+
current = 'o'
|
77 |
+
relation = ''
|
78 |
+
else:
|
79 |
+
if current == 't':
|
80 |
+
subject += ' ' + token
|
81 |
+
elif current == 's':
|
82 |
+
object_ += ' ' + token
|
83 |
+
elif current == 'o':
|
84 |
+
relation += ' ' + token
|
85 |
+
if subject != '' and relation != '' and object_ != '':
|
86 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
87 |
+
return triplets
|
88 |
+
extracted_triplets = extract_triplets(extracted_text[0])
|
89 |
+
print(extracted_triplets)
|
90 |
+
```
|
91 |
+
|
92 |
+
## Model and Tokenizer using transformers
|
93 |
+
|
94 |
+
```python
|
95 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
|
96 |
+
|
97 |
+
def extract_triplets(text):
|
98 |
+
triplets = []
|
99 |
+
relation, subject, relation, object_ = '', '', '', ''
|
100 |
+
text = text.strip()
|
101 |
+
current = 'x'
|
102 |
+
for token in text.replace("<s>", "").replace("<pad>", "").replace("</s>", "").split():
|
103 |
+
if token == "<triplet>":
|
104 |
+
current = 't'
|
105 |
+
if relation != '':
|
106 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
107 |
+
relation = ''
|
108 |
+
subject = ''
|
109 |
+
elif token == "<subj>":
|
110 |
+
current = 's'
|
111 |
+
if relation != '':
|
112 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
113 |
+
object_ = ''
|
114 |
+
elif token == "<obj>":
|
115 |
+
current = 'o'
|
116 |
+
relation = ''
|
117 |
+
else:
|
118 |
+
if current == 't':
|
119 |
+
subject += ' ' + token
|
120 |
+
elif current == 's':
|
121 |
+
object_ += ' ' + token
|
122 |
+
elif current == 'o':
|
123 |
+
relation += ' ' + token
|
124 |
+
if subject != '' and relation != '' and object_ != '':
|
125 |
+
triplets.append({'head': subject.strip(), 'type': relation.strip(),'tail': object_.strip()})
|
126 |
+
return triplets
|
127 |
+
|
128 |
+
# Load model and tokenizer
|
129 |
+
tokenizer = AutoTokenizer.from_pretrained("Babelscape/rebel-large")
|
130 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("Babelscape/rebel-large")
|
131 |
+
gen_kwargs = {
|
132 |
+
"max_length": 256,
|
133 |
+
"length_penalty": 0,
|
134 |
+
"num_beams": 3,
|
135 |
+
"num_return_sequences": 3,
|
136 |
+
}
|
137 |
+
|
138 |
+
# Text to extract triplets from
|
139 |
+
text = 'The Red Hot Chili Peppers were formed in Los Angeles by Kiedis, Flea, guitarist Hillel Slovak and drummer Jack Irons.'
|
140 |
+
|
141 |
+
# Tokenizer text
|
142 |
+
model_inputs = tokenizer(text, max_length=256, padding=True, truncation=True, return_tensors = 'pt')
|
143 |
+
|
144 |
+
# Generate
|
145 |
+
generated_tokens = model.generate(
|
146 |
+
model_inputs["input_ids"].to(model.device),
|
147 |
+
attention_mask=model_inputs["attention_mask"].to(model.device),
|
148 |
+
**gen_kwargs,
|
149 |
+
)
|
150 |
+
|
151 |
+
# Extract text
|
152 |
+
decoded_preds = tokenizer.batch_decode(generated_tokens, skip_special_tokens=False)
|
153 |
+
|
154 |
+
# Extract triplets
|
155 |
+
for idx, sentence in enumerate(decoded_preds):
|
156 |
+
print(f'Prediction triplets sentence {idx}')
|
157 |
+
print(extract_triplets(sentence))
|
158 |
+
```
|