Research On Knowledge Driven Human-machine Active Dialogue Strategy

Conversation in natural language is one of the basic ways of human communication.As the computer has become a powerful tool,humans have a close connection with it.People want to use natural language to give commands and have direct conversations with the computer.Dialogue system has made great progress in recent years,and can carry out coherent and attractive dialogue with human beings.However,the current dialogue mode is still in the initial stage of passive response.How to meet the needs of intelligent humanmachine dialogue system and establish a dialogue system with active response ability is still a huge challenge.In order to enable the dialogue system to have the ability of active dialogue,this thesis studies the management strategy of active dialogue and sequence planning of dialogue objectives.The main contents are as follows:In view of the problem that what the current dialogue system is mainly passive response,it is still unable to carry out active dialogue well.This thesis proposes a knowledge driven man-machine dialogue management strategy actively,to simulate the process of communication between people,interaction can play a leading role on both sides around the subject development and in-depth conversation mode,switch dialogue can be divided into subject and topic into two subtasks,design the personalized dialogue management strategy to implement several rounds of dialogue in the active guidance and the topic.The strategy determines the timing of the system’s active dialogue according to the emotional state of human-computer interaction,and uses the knowledge graph as the background knowledge information to actively search the multi-hop neighbor set of the dialogue entities triggered by the knowledge graph,so as to determine the next interaction content.For topics with negative user emotions,we actively seek new topics through the outward propagation method;for topics with positive user emotions,we use the inward aggregation method to deeply respond to current topics.The experimental results show that this strategy improves the initiative of model dialogue while balancing global dialogue coherence and local topic consistency,providing a new reference for the development of human-machine active dialogue systems.Aiming at the problem that users will dynamically update dialogue targets according to human-computer real-time interaction behavior,but the existing dialogue system has limitations in its adaptability to dynamic dialogue target planning,This thesis proposes a knowledge-driven human-machine dialogue target sequence planning framework,which actively controls the dialogue process through dialogue management,determines the final target considering users’ interests and online feedback,and plans appropriate short-term objectives for natural topic switching.Specifically,the dialogue target sequence is dynamically planned according to the user preference satisfaction and knowledge richness before the dialogue,the completion degree of the user dialogue target is detected in real time during the dialogue,and the final response is selected according to the predicted knowledge,the target to be achieved and the context.When the target changes,the sequence can be re-planned with a new starting point.This framework simulates the behavior of human beings to guide the conversation topic on the knowledge graph,and plans a complete and reasonable target sequence before the conversation,which can guide the user to achieve the conversation goal in a more natural and gentle way.