基于内外知识融合驱动的情感-原因对提取OA

This study addressed the issues of insufficient utilization of intra-text causal clues and the introduction of noise by external knowledge in emotion cause extraction research.This paper proposed a novel method named InExKD-ECPE(internal and external knowledge fusion-driven emotion-cause pair extraction).The method first systematically annotated emotion expe-riencers in the text and constructed experiencer-driven heterogeneous graph network.This network aggregated global semantic information to strengthen the intrinsic connection between emotions and their causes when they shared the same experiencer.Furthermore,it designed a context-aware knowledge fusion module.This module calculated the emotion-sensitive similarity be-tween external knowledge entries and the text based on vad metrics.It then adaptively adjusted the knowledge injection weight by incorporating arousal,thereby effectively filtering out conflicting noise.Experimental results on public datasets show that the proposed method outperforms baseline methods,achieving improvement of 3.4 percentage points on key evaluation met-rics.These results verify the effectiveness of the method for the emotion-cause pair extraction task.