Research On Personalized Video Recommendation Algorithm Based On Knowledge Graph

In recent years,with the rapid development of network communication and the increasing improvement of video device,the entertainment industry represented by network video has achieved unprecedented growth.As a vivid information carrier,video has been widely favored by people.But at the same time,people are facing the increasingly serious problem of "video information overload".In order to help users quickly discover the videos they are interested in among the huge volume of items,the recommendation system emerged to alleviate this issue.The core of the system is recommendation algorithms.The traditional recommendation algorithms represented by collaborative filtering have achieved great success in both research and industrial practice.However,they can't capture users' dynamic preferences,and suffer from the problems of data sparsity,recommendation accuracy and interpretability.In order to overcome the aforementioned challenges,this paper regards the user-item interactions as a behavior sequence,and captures users' dynamic preferences by modeling the behavior sequence.Furthermore,this paper integrates the knowledge graph into the recommendation algorithm,and uses the semantic information of the knowledge graph to enrich the representation of items,so as to solve the issues of data sparsity,recommendation accuracy and interpretability.The content of this paper consists of four parts:(1)To solve the problem that traditional recommendation algorithms can't capture users' dynamic preferences,a sequential recommendation model based on user-item graph structure is proposed.This model characterizes user-item interactions as a bipartite graph structure,and then uses graph neural network to learn representation of node,so as to obtain a more effective representation of users and items.Finally,the self-attention network is employed to model user's behavior sequence.(2)In order to vectorize the knowledge graph and make better use of its semantic information,a knowledge graph embedding model based on graph convolutional network is proposed.It combines the encoder of Comp GCN with the decoder of SACN,and leverages the Bi-Interaction aggregation mechanism to improve the original aggregation of Comp GCN.(3)To tackle the problems of data sparsity,recommendation accuracy and interpretability,a sequential recommendation model based on knowledge graph embedding is proposed.This model integrates the knowledge graph embedding obtained in(2)into the recommendation model proposed in(1)and enrichs the item representation leveraging the semantic information from the knowledge graph.(4)A video recommendation system is designed and implemented based on the recommendation model proposed in(3),and its personalized recommendation function to users is demonstrated through experimental test.