Datasets:

Modalities:
Text
Formats:
csv
Size:
< 1K
Libraries:
Datasets
pandas
License:
stellawang196 commited on
Commit
c08da6a
·
verified ·
1 Parent(s): 50078ac

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -17,7 +17,7 @@ Accurate and consistent definitions of terms in financial regulations are import
17
  | Securities and Exchanges | 13 | SEC |
18
 
19
  # Metrics
20
- BERT score is used.
21
 
22
  # Related tasks
23
  Regulations Challenge at COLING 2025: https://coling2025regulations.thefin.ai/home
 
17
  | Securities and Exchanges | 13 | SEC |
18
 
19
  # Metrics
20
+ We use BERTScore to evaluate the generations, measuring the semantic similarity between the LLM-generated answers and the provided definitions. It converts both reference and candidate sentences into contextual embeddings using transformer-based models like BERT. The semantic alignment is quantified through cosine similarity between these embeddings, providing a metric of how closely the model's responses match the expected semantic content of the definitions.
21
 
22
  # Related tasks
23
  Regulations Challenge at COLING 2025: https://coling2025regulations.thefin.ai/home