On Recommendation Methods For Attributed Graph Through Deep Learning

With the rapid development of the Internet,various kinds of information appear in people's vision,the recommendation system becomes more and more important.The traditional recommendation algorithm suffers from problems such as cold start,data sparsity and so on.Heterogeneous information network,as a widely existing network structure in practical applications,is rich in structural information that is helpful to solve the above problems.Heterogeneous attribute graph are more powerful as heterogeneous information networks with node attributes.Due to the rapid increase of computer performance in recent years,the rapid development of deep learning,high complexity and high non-convex heterogeneous attributed graph can also be learned though deep learning.How to use the topological structure and attribute information in heterogeneous attribute graph to further improve the performance of the recommended system is the main content of this paper.It can be subdivided into two parts,that is,how to learn heterogeneous attribute graph embedding,and how to use graph embedding to recommend.For the problem of embedding on heterogeneous attribute graph,this paper proposes a graph embedding algorithm AAGE(Autoencoder based Attributed Graph Embedding)based on the fusion of topological features and node attribute features,and proposes the concept of inverse occurrence weight.Based on auto-encoder,AAGE algorithm uses the embedded layer to deal with structure information and attribute information,and weighting the links and attributes by inverse occurrence weight.Also,this paper designs the loss function based on first-order and second-order similarity,and gives a detailed derivation formula.Finally,the validity of the algorithm is verified through experiments.As for recommendation for graph embedding,since the network structure is mostly bipartite graph in real world,this paper proposed ENBI(Embedding Network-Based Inference)algorithm to solve the problem of recommendation based on graph embedding.The ENBI algorithm is essentially an extension of the NBI algorithm in the vector dimension.Through the resource allocation back and forth between two types of nodes in the bipartite graph,the NBI algorithm obtains a relative score between the nodes and generates a recommendation list by using the scores.This paper proposed the design of a recommend system based on the AAGE algorithm and ENBI algorithm,and describes details of its implementation.The experiment shows that the system have achieved a good improvement.