Research On Personalized Recommendation Based On Graph Representation Learning

The development of information technology and the Internet industry is gradually changing people's way of life.Various platforms have launched intelligent solutions.As a result,the derived data is huge and rich.Among them,graph data is widely used,such as knowledge graph,social networks,etc.But these graph data are usually complex in structure and large in scale,which challenges users' information acquisition.Although the traditional recommendation algorithm can alleviate the above difficulties,it can not effectively deal with the graph data information,can not capture the user's personalized preferences,and will lead to data sparsity and cold start problems.In view of the above problems,this paper adopts the personalized recommendation technology based on graph representation learning.According to the graph structure of knowledge graph,it combines the corresponding graph representation learning method to learn the feature representation of users and items,and based on this feature representation to recommend items for users.The main contributions of this paper are as follows:In this paper,a personalized recommendation model based on user-end and item-end knowledge graph(UIKG)is proposed.The model mines the association attribute information of user and item in their knowledge graph(KG),and effectively captures the association between user's personalized preference and item through joint learning.In the specific operation steps,the method based on graph convolution neural network is used to learn the user representation vector from the user knowledge graph,and then the user representation vector is introduced into the item knowledge graph to jointly learn the item representation vector,so as to realize the seamless unification of the user KG and the item KG.Finally,the preference probability of the user to the item is obtained through the multi-layer perceptron,Thus,the higher-order structure information and semantic information in KG can be mined more effectively to capture the user's personalized preferences.Many KG-based recommendation models,including UIKG model,do not consider the user's trajectory sequence information.However,many of the models based on sequence recommendation do not consider the context information of the interaction between the user and the project,so they are not suitable This paper proposes a multi context sequence recommendation model(MCSR)based on KG to solve the lack of user trajectory sequence information in KG recommendation and the lack of context information between user and project interaction.The model is mainly divided into two modules: UT2 vec and CKG2 vec.The sequence module UT2 vec is used to learn the user's trajectory sequence information,and the user's corresponding behavior trajectory is mapped to KG one by one.The CKG2 vec module is used to learn the context information of user project interaction.Finally,the feature representation of user and project is integrated,and the user's personalized preferences are deeply modeled through multi-layer perceptron.