Empowering Small Language Models with Factual Hallucination-Aware Reasoning for Financial Classification
This work addresses the issue of factual hallucinations in SLMs for financial classification, which is an incremental improvement aimed at making SLMs more trustworthy and effective in finance.
The paper tackled the problem of factual hallucinations in small language models (SLMs) for financial classification by proposing a three-step pipeline (AAAI) that detects and adapts to these errors, resulting in enhanced classification performance as shown in experiments on three SLMs.
Small language models (SLMs) are increasingly used for financial classification due to their fast inference and local deployability. However, compared with large language models, SLMs are more prone to factual hallucinations in reasoning and exhibit weaker classification performance. This raises a natural question: Can mitigating factual hallucinations improve SLMs' financial classification? To address this, we propose a three-step pipeline named AAAI (Association Identification, Automated Detection, and Adaptive Inference). Experiments on three representative SLMs reveal that: (1) factual hallucinations are positively correlated with misclassifications; (2) encoder-based verifiers effectively detect factual hallucinations; and (3) incorporating feedback on factual errors enables SLMs' adaptive inference that enhances classification performance. We hope this pipeline contributes to trustworthy and effective applications of SLMs in finance.