Dataset · FutureMa
EvasionBench benchmark for evasive answers
FutureMa/EvasionBench
16,726 Q&A pairs from earnings calls with three evasion levels for classification.
What's inside
EvasionBench is a benchmark for detecting evasive answers in earnings call Q&A. It contains 16,726 Q&A pairs from financial call transcripts labeled with three evasion levels. Labels were generated by Eva-4B-V2, a fine-tuned classifier. Language is English.
What you can do with it
- Train an evasion response classifier
- Evaluate models on financial Q&A
- Check the labeling quality of Eva-4B-V2
- Tasks
- Text classification
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 402
- Likes
- 111
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("FutureMa/EvasionBench")