Research On Human-computer Interaction Content Recommendation Method Based On Knowledge Graph

Since the communication between people is content and coherent,in the process of human-computer interaction,it is hoped that the computer's replies are also content and coherent.How to enable computers to recognize,understand,and generate quality content and coherent responses like humans has received widespread attention.In order to enhance the quality and continuity of the robot's response content,this thesis aims to explore the content recommendation methods of robots in two interactive systems,open domain and closed domain,respectively.The main research contents are as follows:Aiming at the problems of lack of robot background knowledge and low response coherence in the current open domain human-computer interaction system,a content recommendation method based on the knowledge graph ripple network is proposed.In order to achieve better content and a more coherent human-computer interaction system,the model method simulates the process of real human-to-human communication.First,the human-computer interaction emotion friendliness degree is obtained by calculating the evaluated emotion value and the emotion confidence of human-computer interaction.Then,the external knowledge graph is introduced as the background knowledge of the robot,and the dialogue entity is embedded in the knowledge graph ripple network to obtain the entity content of participants potential interest.Finally,comprehensive consideration of emotional friendliness and content friendliness will offer a robot reply.The experimental results suggest that,compared with the comparison model methods such as the MECs model,a robot with background knowledge and emotional measurement can effectively improve its emotional friendliness and coherence when performing human-computer interaction.Aiming at the problem of low quality and low consistency of robot response content in the current closed-domain human-computer interaction system,a knowledge-perceived a content recommendation method for customer service robots is proposed.In order to achieve higher accuracy and more coherent content response,the model method obtains better results by perceiving the deep learning network based on the knowledge of the attention mechanism.First,the system performs knowledge extraction based on the content of the participants conversation,so as to obtain the conversation entity collection of the participants.After obtaining the conversation entities,which are linked to form the word link entity,the system will perform a first-order external expansion for each entity in the collection,and the word expansion entity is formed.Secondly,the obtained participant dialogue is vectorized,and the vector set of the participant's greatest focus is obtained through the knowledge perception deep learning network.Finally,based on the attention mechanism,the participant's attention focus is dynamically obtained so that the best content response can be offered.The experimental results show that,compared with the comparative model methods such as the DISAN model,the interactive model with knowledge perception can effectively improve the quality and coherence of the response content during human-computer interaction.