Improving unified information extraction in Chinese mental health domain with instruction-tuned LLMs and type-verification component

Zijie Cai, Hui Fang, Jianhua Liu, Ge Xu, Yunfei Long, Yin Guan, Tianci Ke

Abstract: Highlights•Unified mental health IE datasets established.•Enhanced extraction via multi-instruction LLM instruction-tuning and data augmentation.•Small Parameter Models Boost the performance of IE and type-verification efficiency.•Innovated a simplified method for constructing Type-verification dataset samples.