-
Notifications
You must be signed in to change notification settings - Fork 0
/
CrossMatching.py
59 lines (51 loc) · 2.4 KB
/
CrossMatching.py
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
from sentence_transformers import SentenceTransformer, util
from dotenv import load_dotenv
import os
import torch
import json
load_dotenv('.env')
# Load Limit Score constant from env
limit_score = float(os.environ['LIMIT_SCORE'])
input_cn_file_path = os.environ['OUT_CN_FILE_PATH']
input_en_file_path = os.environ['OUT_EN_FILE_PATH']
result_file_path = os.environ['RESULT_JSON_FILE_PATH']
# save model in current directory
model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2', device='cpu', cache_folder='./')
# save model in models folder (you need to create the folder on your own beforehand)
# model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2', device='cpu', cache_folder='./models/')
print("Read Input Files ...")
# Read queries from file and split by line breaks
with open(input_en_file_path, "r", encoding='utf-8') as query_file:
queries = query_file.read().split("\n\n")
# Read corpus from file and split by line breaks
with open(input_cn_file_path, "r", encoding='utf-8') as corpus_file:
corpus = corpus_file.read().split("\n\n")
def similarity_score(query, document):
# Calculate cosine similarity between query and document
query_embedding = model.encode(query, convert_to_tensor=True)
document_embedding = model.encode(document, convert_to_tensor=True)
cosine_similarity = util.pytorch_cos_sim(query_embedding, document_embedding)
return cosine_similarity
def get_matching_documents(queries, corpus):
matching_documents = []
querylist_length = len(queries)
documentlist_length = len(corpus)
for query in queries:
query_embedding = model.encode(query, convert_to_tensor=True)
for document in corpus:
document_embedding = model.encode(document, convert_to_tensor=True)
cosine_similarity = util.pytorch_cos_sim(query_embedding, document_embedding)
if cosine_similarity > limit_score:
result = {
"query": query,
"document": document,
"score": cosine_similarity
}
matching_documents.append(result)
return matching_documents
print("Calculate Similarity Scores ...")
matching_documents = get_matching_documents(queries, corpus)
print("Save Results ...")
with open(result_file_path, 'w', encoding="utf-8") as file:
json.dump(matching_documents, file, ensure_ascii=False, indent=4)
print("Done!")