Research shows that LLM-based emotion-cause pair extraction performs better using pair-level judgement than dialogue-level generation. While models struggle to discover entire sets of relations, they can recognize 92.7% to 98.1% of relations when queried with explicit, individual pairs.
HOW THIS AFFECTS YOU
●
researcherYou should shift from generative formulations to judgement-based architectures to improve extraction recall and precision in conversational AI.