CDEA: Causality-Driven Dialogue Emotion Analysis via LLM
Year: 2025
Type: article
Abstract: As human-machine dialogue technology rapidly advances, the importance of sentiment analysis in dialogue systems has become increasingly prominent. However, due to the subjectivity of emotions, existing sentiment analysis methods based on deep learning face reliability issues. To address, this paper proposes a model based on explicit reasoning about emotional causes, which extracts reasoning path information from structured commonsense knowledge to capture the causal relationships between historical and target sentences. Furthermore, by utilizing a pre-trained language model, emotional cause-oriented contexts are constructed through prompts to reason and integrate emotional causes,thereby improving the accurate classification of emotion categories. In addition, to address the lack of commonsense knowledge support in current methods, this paper introduces an emotion cause analysis method that incorporates large language models. By building an experience database and dynamically retrieving examples similar to the target sentence, the commonsense knowledge and generalization capabilities of LLM are leveraged to enhance sentiment classification performance. To further improve the adaptability of the model, a task is designed to generate semantic interpretations of emotion categories, and the LLM is fine-tuned accordingly. Experimental results show that the proposed method performs excellently on three public datasets, significantly improving the accuracy and reliability of dialogue sentiment analysis.
Source: Symmetry
DOI: 10.3390/sym17040489
Authors: Xue Zhang, Mingjiang Wang, Xuyi Zhuang, Xiao Zeng, Qiang Li
Institution: Harbin Institute of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W4408833052