Research On The Algorithm Of Knowledge-driven Personalized Dialogue System

With the development of deep learning technology and the explosive production of Internet data,the establishment of a human-computer interactive dialogue system has gradually become possible.The dialogue system has huge potential commercial value and social value,and it has a wide range of application scenarios,such as customer service robots and chat assistants.In the near future,the application of intelligent dialogue systems will become more and more common.Therefore,it is very necessary to study dialogue systems in depth.This paper mainly studies the generative dialogue system in the open field.The generative dialogue system is mainly based on the "encoder-decoder"framework,and the model is trained end-to-end on large-scale dialogue data.This article focus on the three important factors that affect the dialogue,namely dialogue context,background knowledge,and persona,and try to improve the quality of the responses generated by the dialogue model.The research content of this article mainly includes the following two aspects:(1)The dialogue context contains information such as chat background and themes,but the dialougue system is not enough to generate informationrich replies based on the dialogue context.It also needs to combine background knowledge related to the dialogue context to make up for the semantic gap between the response and the input message.Therefore,This paper proposes a context-sensitive reply generation model based on background knowledge,which combines the dialogue context and background knowledge information in a unified framework to generate rich responses that conform to the context.This paper uses a static graph attention mechanism to encode multiple knowledge triples as a whole graph,and merge them with word vectors to enrich and enhance the semantic information of words,and promote the dialogue system to better understand the dialogue context.In addition,in the decoding stage,the graph-level and triple-level attention mechanisms are used to select and read the appropriate knowledge triples,and the entities of the triples are copied to the response.(2)The speaker is the core of the dialogue process.If the personalized information of the speaker is ignored,the responses generated by the dialogue model will have inconsistencies in language style,reply semantics,and hobbies.In order to better combine personalized information,this paper proposes a personalized dialogue model based on bidirectional attention flow.In this paper,the bidirectional attention flow is used to realize the interaction between the dialogue history vector and the user portrait vector,and obtain the contextual representation of the perceived personalized information,and then promote the dialogue system to generate a consistent personalized response.