Research And Implementation Of Multi-turn Chatbot Based On Knowledge Graph

Chatbot is a kind of chatting system that generates real-time natural language feedback according to the user's natural language input.Because of its rich application scenarios and potential commercial value,the chatbots which can communicate with people smoothly and naturally has always been the focus in the field of artificial intelligence.However,due to the lack of knowledge and the inability to better understand the logical relationship between things,chatbots generally have the problem of insufficient knowledge reasoning ability;at the same time,because they can't review and use historical information as efficiently as human brain,chatbots tend to deviate from the topic of chatting in conversation,and their ability of multi-turn conversation is weak.Knowledge graph is a kind of graph data structure used to record knowledge and reveal the logical relationship between things.Using effective algorithms can also achieve more complex logical reasoning operation,which is an effective way to improve the performance of chatbots.This paper focuses on the research of multi-turn chatbot based on knowledge graph and obtains the following achievements:(1)In order to solve the problem that chatbots are lack of knowledge reasoning ability and easy to generate wrong answers,a deep learning model KE-Seq2Seq,which can effectively use knowledge graph,is proposed.Through attention calculation,the graph knowledge related to user input message is integrated into the encoder of the model(encoder knowledge fusion mechanism),so as to improve the knowledge reasoning ability of the model.The results show that the fluency,accuracy,recall and F1 value of the model are significantly improved.(2)In order to solve the problem that chatbots are weak in multi-turn conversation and easy to deviate from the theme,this paper proposes a deep learning model KD-Seq2Seq,which can effectively use the historical map information.Through dual attention computing,the map knowledge related to historical dialogue is integrated into the decoder of the model(decoder knowledge fusion mechanism),which effectively enhances the ability of multi round dialogue with users.The results show that the fluency,accuracy,recall,F1 value and rationality of the model are significantly improved.(3)The encoder and decoder knowledge fusion mechanisms are applied together to a joint enhancement model of knowledge reasoning and multi-turn conversation capability--KED-Seq2Seq,which is better than the single enhancement model.Based on the joint enhancement model,the web service of chatbot is developed,which realizes a closed-loop message,complete process,friendly interaction and efficient and available chat robot system.