Research On Dialogue Generation Technology Based On Topic Enhancement And Semantic Consistency

The main goal of human-computer dialogue is to hope that the machine can have a smooth and natural dialogue with human beings,so as to facilitate human life and meet human needs.In order to enhance the quality of dialogue and make the machine more like human beings,a variety of man-machine dialogue systems have emerged.However,in the open domain,how to make the machine clearly understand the human intention and clear the goal of dialogue is still a difficult problem.Therefore,this paper focuses on dialogue generation in open domain,which aims to generate more logical,contextual,thematic and diverse reply content.The main work of this paper is as follows:(1)Research on single round dialogue generation technology based on explicit topic enhancement.Although some of the existing dialogue generation methods focus on topic relevance,it is still a difficult problem for researchers to explore and solve how to make the reply content highly related to the topic of the source sentence.In the open domain,the existing dialogue system does not pay enough attention to the topic when generating sentences,and the subject word sequence extracted from the text does not pay attention to the information screening when adding into the model,which is easy to cause large semantic errors.To solve this problem,this paper constructs a single round dialogue generation model based on explicit topic enhancement.In the process of dialogue,appropriate word vectors are explored in various ways,focusing on the consistency of semantic information and topic information between two sentences,so as to generate high-quality dialogue content with semantic coherence and topic consistency.We do a series of comparative experiments on the model,and the results show that the model has a certain improvement in the quality of generated dialogue.(2)Research on multi round dialogue generation technology based on information interaction and topic expansion.On the premise of improving the quality of single round dialogue,we focus on the research of multi round dialogue.Although the existing multi round dialogues pay attention to the cohesion between the reply content and the context,they i gnore the expansion of the topic,resulting in the tendency to generate sentences that end the topic,such as "mm-hmm,OK","I see","OK,OK",which are not suitable for daily small talk conversations.To solve these problems,this paper proposes a multi round dialogue generation model based on information interaction and topic expansion.The model constructs the information interaction between the reply content and the context,and ensures the semantic consistency between the reply content and the context.The topic expansion module is designed to expand the topic in the dialogue process,so that the human-computer communication can be continued,and at the same time,the topic expansion module is also designed Based on the idea of CVAE,a new priori network and reasoning network are constructed to improve the diversity of reply content.In addition,we let the model generate the content in any position,and also need to be able to connect with the context,which further enhances the understanding ability of the model.Finally,experiments are carried out on LCCC datasets.The experimental results based on automatic evaluation and manual evaluation show that the model has certain advantages.