Personalized Recommendation Based On Behavior Sequence And Knowledge Enhancement

The recommendation system is a tool to assist users in filtering information,and provide users with personalized recommendation services of interest by analyzing users’ behavior and content.Knowledge graph is an effective means to efficiently manage and visualize a large amount of data,which can improve the intelligence level of information management.Knowledge graph has gained great attention in the recommendation system as an auxiliary tool for expanding information,which provides more accurate recommendations by mining the relationship between entities and linking implicit correlations between items and users.The integration of knowledge graph into recommendation systems is an important direction for future research,which can improve the interpretability of recommendation systems and help users understand the potential reasons behind recommendation results.At the same time,this also helps to increase user satisfaction with the recommendation results.In most recommender system models,user interaction behavior is regarded as an ordered sequence,and users can extract dynamic behavior sequence features through the self-attention mechanism,but do not consider the time interval between each interaction.Entities and items are interrelated and highly interconnected in the Knowledge Graph and Recommendations module respectively.At the same time,in the knowledge graph module,the propagation path of the entity is random,without considering the historical preferences of the user.Aiming at the above two problems,this paper proposes a TPMKR model,a multi-task feature learning recommendation model based on the fusion of behavioral sequence time intervals and entity knowledge enhancement.The timestamp of the interaction is modeled in the behavior sequence,and the self-attention mechanism is introduced in the knowledge graph module to enhance the entity knowledge,and the optimal entity propagation path is obtained through weighted calculation,and the final recommendation is completed through the cross unit of the model.In the experiment,the AUC,ACC,Precision,and Recall indicators were used to evaluate the proposed model and mainstream benchmark model on the real dataset Movie Lens-1M.Experimental results show that the proposed model outperforms the mainstream benchmark model.This paper stores the movie data structured triplet in Neo4 J to build a knowledge graph in the film field,and completes the design and implementation of the movie recommendation system based on the TPMKR model,uses Spring Boot and Vue.js to complete the front-end and back-end development,and MySQL stores the data.The purpose of this recommendation system is to verify the advantages of the model based on basic requirements such as various functions,the interface for displaying movies and the recommendation page.