Empathetic Dialogue Response Generation Research Based On Commonsense Reasoning

Dialogue generation system is an important research direction in the field of natural language processing,which has been widely applied in areas such as online education platforms and customer service.Dialogue systems with empathy can deeply understand the information conveyed by users and perceive their emotional states,thus making empathetic responses,which has drawn extensive attention from researchers.At present,the proposed empathetic dialogue generation models mainly rely on identifying and modeling the user’s current emotional state to generate empathetic responses.However,they often overlook the consideration of commonsense knowledge and logical reasoning abilities,making it difficult to understand the deep meaning behind user utterances.In addition,the empathetic responses they generate are often unpredictable and have poor controllability,leading to generic or inappropriate responses.To address these issues,this paper endows the empathetic dialogue model with common sense reasoning ability,and based on this,models the expected emotion to be conveyed in the response(i.e.,anticipated emotion)and the questioning strategy used in the response.This work can help generate more empathetic responses in dialogue systems.The main research content of this paper is as follows:(1)A reinforcement learning-based empathetic dialogue response generation model is proposed that incorporates commonsense reasoning and anticipated emotion in generating empathetic responses.Specifically,the model first performs commonsense reasoning on the user input based on a commonsense knowledge graph to obtain relevant information for generating empathetic responses.Then,the obtained information is used to assist the model in generating empathetic responses.Finally,at the sentence level,multiple reward signals,including the anticipated emotion,are extracted to guide the model in generating empathetic responses that express emotions appropriately and with high quality.Additionally,the sentence-level rewards effectively alleviate the exposure bias problem during training.Experimental results demonstrate that the proposed method outperforms current mainstream baseline models in both automatic and human evaluations,and the generated responses have better controllability.(2)A deep learning-based empathetic dialogue response generation model is proposed that incorporates commonsense reasoning and questioning strategies in generating empathetic responses.The model not only has the ability to perform common sense reasoning but also considers questioning strategies during the response generation process to improve the controllability of the generated responses.Specifically,the model first uses a commonsense knowledge graph to infer implicit information from the user input,and then combines the obtained additional information with the user input to perceive the user’s current emotional state.Finally,a multi-task learning approach is used to construct three loss functions,which are used for emotion recognition,questioning strategy,and response generation,respectively.Testing on a public dataset shows that the proposed model has good emotion recognition capabilities and performs excellently in perplexity.