Research On Recommendation System Based On Network Representation Learning

With the rapid development of the Internet,the amount of data we face is growing exponentially.Users are faced with serious information overload,and it is difficult to find the required content from all resources in a short time.The purpose of recommendation system is to provide users with personalized recommendation items according to their preferences to improve the user experience.However,the recommendation system still has the problems of data sparsity and cold start.Network is a common data form in the world,which has more structural information than text data.Knowledge graph is a special heterogeneous network,which has the advantages of accuracy,diversity and interpretability.As auxiliary information in the recommendation system,it can mine more accurate and higher-order information,so as to enhance the expression ability of the recommendation model.Therefore,this paper attempts to use the common network structure and knowledge graph to provide more sufficient information for the recommendation system to enhance the representation,so as to improve the effect of the recommendation system.This paper proposes to use the network representation learning method and knowledge graph in the recommendation system,and use the rich structural semantic information and high-order connectivity of the network and knowledge graph to reduce the data sparsity and cold start problems of the recommendation system.According to the data characteristics of different scenarios,this paper reconstructs and supplements the data by using the general network structure and the form of knowledge graph,and then combines the network representation learning and knowledge graph embedding methods with the recommendation task to mine more sufficient node characteristics.Integrating the method of ordinary network representation learning into the recommendation system has greater universality and expansibility.The method of embedding the knowledge graph into the recommendation system needs to construct the knowledge graph,but the structure and semantic advantages of the knowledge graph provide high-quality node features for the recommendation model and effectively improve the user experience of the recommendation system.The main research contents and contributions of this paper are as follows:(1)By analyzing the data of different scenarios of the recommendation system,this paper constructs the user project interaction data as a bipartite graph and the project attribute information as a knowledge graph.On this basis,the graph is improved.Firstly,the entity alignment knowledge fusion is carried out between the bipartite graph and the attribute knowledge graph,and a connected graph containing both user project interaction information and project attribute information is obtained.Secondly,connect the edges according to the similarity between nodes and the similarity between projects,so as to better describe the user portrait and project characteristics,so as to enhance the high-order connectivity of the network structure.(2)This paper proposes a type aware graph convolution network(type aware GCN)recommendation algorithm.This algorithm introduces the network structure into the recommendation system.Firstly,type aware neighbor sampling and aggregation operations are used to learn the specific type of neighborhood representation.Then,the attention mechanism is used to distinguish the importance of different node types.Through experiments on public data sets,compared with the current mainstream recommendation algorithms,it is verified that the introduction of network structure and node classification perception can improve the performance of the recommendation system.(3)This paper proposes a knowledge graph recommendation algorithm based on multi granularity aggregation(MAKR).This algorithm proposes a new aggregator based on graph neural network,which comprehensively considers type aware attention,fine-grained Transformer and coarse-grained FM to aggregate neighbor information,and depicts the representation of users and items from multiple granularities.Through experiments on three public data sets,it is verified that aggregating neighbor information from multi granularity can improve the performance of the recommendation system.