PreranTej/roberta-cognitive-bias-multitask

RoBERTa Cognitive Bias Multi-Task Model

Detects cognitive biases in STEM literature and extracts the biased span from text.

Labels

0: Neutral | 1: Anchoring Bias | 2: Bandwagon Bias 3: Confirmation Bias | 4: Framing Bias | 5: Emotional Appeal 6: Appeal to Authority | 7: False Cause & Certainty 8: Hasty Generalization

Performance

• —Macro F1: 0.864
• —Span Match Rate: 98% on validation set

Usage

Load with the custom modeling file included in this repo.