Improving Empathetic Response Generation by Emotion Recognition and Information Filtration

Guodong Shi,Hongxu Hou,Wei Chen,Shuo Sun,Yuan Zhao,Jipeng Ma
DOI: https://doi.org/10.1007/978-981-97-5669-8_14
2024-01-01
Abstract:Empathetic response generation is commonly regarded as the result of harmony and interaction between affection and cognition. Thus, it is necessary to identify emotions and consider the situation in empathetic response generation. Although recent methods enhance the understanding of the user's situation by incorporating commonsense knowledge, we believe that previous attempts are incomplete because they neglect three main issues: firstly, they overlook the role of commonsense knowledge in emotion recognition. Secondly, inconsistency between emotions and information can lead to responses that do not match the actual situation. Finally, splicing dialogue context and commonsense knowledge complicates the model's understanding and processing. To solve the above problems, we propose a novel method that consists of three key sections: emotion recognition, information filtration and response generation. Specifically, this paper introduces a novel emotion recognition method, conducts the Context Filter and the Knowledge Filter, and uses a multi-head attention mechanism to handle the spliced sequence of dialogue context and commonsense knowledge. Extensive empirical results with EmpatheticDialogues demonstrate that our approach outperforms existing baselines to recognize emotions more accurately and generate more empathetic responses.
What problem does this paper attempt to address?