QADYNAMICS: Training Dynamics-Driven Synthetic QA Diagnostic for Zero-Shot Commonsense Question Answering

TL;DR

QADYNAMICS is a training dynamics-driven framework that refines synthetic QA data by analyzing training behavior, removing noise and false negatives, thereby improving zero-shot commonsense QA performance with less data.

Contribution

It introduces a novel training dynamics-based method for QA diagnostics and refinement, reducing noise in synthetic data and enhancing model generalization in zero-shot commonsense QA.

Findings

Outperforms baselines using only 33% of synthetic data

Significantly improves QA quality according to expert evaluations

Effective even with large language models like ChatGPT

Abstract

Zero-shot commonsense Question-Answering (QA) requires models to reason about general situations beyond specific benchmarks. State-of-the-art approaches fine-tune language models on QA pairs constructed from CommonSense Knowledge Bases (CSKBs) to equip the models with more commonsense knowledge in a QA context. However, current QA synthesis protocols may introduce noise from the CSKBs and generate ungrammatical questions and false negative options, which impede the model's ability to generalize. To address these issues, we propose QADYNAMICS, a training dynamics-driven framework for QA diagnostics and refinement. Our approach analyzes the training dynamics of each QA pair at both the question level and option level, discarding machine-detectable artifacts by removing uninformative QA pairs and mislabeled or false-negative options. Extensive experiments demonstrate the effectiveness of our approach, which outperforms all baselines while using only 33% of the synthetic data, even including LLMs such as ChatGPT. Moreover, expert evaluations confirm that our framework significantly improves the quality of QA synthesis. Our codes and model checkpoints are available at https://github.com/HKUST-KnowComp/QaDynamics.