LLM conversation analysis: Pair judgment outperforms generation

A new paper from arXiv explores the effectiveness of different approaches for emotion-cause pair extraction in conversations using large language models (LLMs). The research found that formulating the task as a pair-level judgment, where the LLM assesses individual candidate pairs, significantly outperforms dialogue-level generation methods. This is because LLMs are better at recognizing existing relations when explicitly queried rather than discovering and generating complete sets of relations. The study introduces an auxiliary retriever to improve performance by re-examining ambiguous cases, leading to consistent F1 score improvements. AI

IMPACT Suggests task decomposition and candidate scope are critical for effective LLM use in relation extraction.

RANK_REASON Academic paper on LLM capabilities for a specific NLP task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• ECPEC

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →