Research On Knowledge-driven Generative Conversation System

The rise of a new generation of artificial intelligence technology represented by deep learning has promoted the vigorous development of natural language processing technol-ogy.As a typical application of natural language processing technology,human-computer conversation system is the current research hotspot in academia and industry.The tradi-tional generative dialogue system is limited by lack of external knowledge,there are a se-ries of problems such as single sentence generation and poor dialogue interaction ability.In recent years,knowledge-driven artificial intelligence has received extensive attention from scholars.This thesis introduces knowledge into the generative conversation system and improves the conversation quality.The specific research contents of this thesis are as follows:1.A generative dialogue model based on knowledge reasoning is proposed.Aiming at the problem of utilization of knowledge entities in the generation process,the model designs a knowledge selector based on the attention mechanism and a knowledge entity reasoner based on the attention mechanism.The knowledge selector based on the atten-tion mechanism strengthens the context semantic coding? The Knowledge entity reasoner based on attention mechanism increases knowledge entity prediction accuracy through global knowledge information fusion and knowledge entity inference process,and can help to enrich the answer.Aiming at the problem of different structural data in the exist-ing data set,the model designs a heterogeneous knowledge processor,so that the model can support the input of structured knowledge and continuous knowledge,and eliminate the spatial inconsistency of semantic encoding through the conversion network? Also de-sign a decoder based on knowledge reasoning.Finally,a comparative experiment on the Ducov data set confirms that the model can make the generated sentences contain external knowledge and improve the quality of the generation.2.The knowledge pointer generation dialogue model is proposed to improve the problem of low knowledge coverage of generated sentence knowledge in the generative dialogue model based on knowledge reasoning.The model is improved in two aspects.The first is to address the problem that the attention mechanism cannot fully capture the connection between long context and knowledge.Combining the Bidirectional at-tention mechanism improves the global knowledge information fusion vector and adds fine-grained semantic features to enhance Knowledge entity prediction accuracy rate.The second is to improve the pointer-generator mechanism.We have designed the generation of the probability distribution of input words,the generation of the probability distribu-tion of knowledge entity words,the generation of the probability distribution of word lists,and the corresponding decoder and loss function.Through experimental analysis,it is confirmed that the knowledge pointer network can effectively solve the problem of low knowledge coverage,make the generated information richer in answer,and at the same time improve the OOV problem,further improve the quality of the generated response.3.Based on the above work,the thesis designs and implements a human-machine dialogue system for the entertainment movie field.Tests have shown that the model pro-posed in this thesis has practical effects in terms of conversational richness and improved sentence generation quality,which lays the foundation for further research.