Research And Application On Multi-relation Network Embedding

A multi-relationship network structure is a data structure that contains many different association information between things and things.The nodes and edges in this structure usually represent things and multiple relationships respectively.Compared with network structure with only one single type of relation,a multirelation network structure can express more complex relationships between entities.And therefore,multi-relation network has a wider range of applications,such as bioinformatics,linguistics,and social community.Most of the existing graph embedding methods are designed for problems where there is only one single type of relationship between nodes,rather than multi-relation network structures.However,in real life,more data have many different kinds of relationships.As the complexity of relational data increases,researchers are now beginning to realize the importance of multi-relation network embedding.In this paper,we focus on two specific types of multi-relation network.First one is one single network with multi-relation.This type of network contains more than one type of node.And the difference types between the nodes leads to a consequent difference type of edges.The other one type of multi-relation network is multi-view network structure.In a multi-view network structure,the types of edges on the graph are not determined by the types of the two connected nodes,but by the original data.And there may exist more than one type of relationship between the nodes.To solve the above problems,this paper optimizes the graph embedding algorithm in the traditional single-relation network to adapt to the multi-relation network structure,so that the information of relationship types in the graph can be better utilized.For the one single network with multi-relation structure,we present our model on the task of predicting pathogenic genes.We use known correlations between the entities,which may be from different sets,to build a biological heterogeneous network and propose a new network embedded representation algorithm to calculate the correlation between disease and genes,using the correlation score to predict pathogenic genes.Then,we conduct several experiments to compare our method to other state-of-the-art methods.The results reveal that our method achieves better performance than the traditional methods.And for the multi-view network structure,we propose a multi-view graph autoencoder method.This method can fuse the information of the multiple views into a unified embedding space without losing their distinctive properties.We conduct experiment of link prediction on three different multi-view graph datasets.And the result proves that our proposed method performs better than the current graph embedding method either designed for single view or multi-view.