Research on the Design of Intelligent Chatbot Based on Deep Learning

Abstract

At present, chat robots still have the problems of insufficient utilization of multi-modal information, weak emotion expression ability, and lack of model fusion mechanism. With the application of deep learning in the fields of natural language understanding, word vector representation, machine translation, sentiment analysis, and Chinese word segmentation, people have begun to study the key technologies of chat robots and apply deep learning to chat robots. In view of the problem that the traditional chat robot Seq2Seq model training will produce too safe, common, and repetitive answers to the speech corpus, which affects the interactive experience with users, this paper combines the idea of mutual information to improve the objective function of the model so that the model can produce richer and more diverse answers and give users a better experience. Experiments show that the model has improved many indicators compared with the single-modal model in the microblog data set experiment.