Research On Network Embedding Link Prediction Algorithm Based On Fusion Topology And Node Attributes

Network science as a new subject,has developed rapidly.Link prediction has important research significant in many fields,such as friend recommendation in social network and gene regulation.The network generally contains two types of information: the network topology formed by the connection relationship between nodes and the attributes of the node itself.The network topology information can easily be acquired and calculated.The link prediction algorithm based on network topology only considers some topological properties,it causes the specific structural indicators hard to have high performances in all networks.It is difficult to give a generality structural similarity indicator.At the same time,in a lot of networks,there are rich node attribute information besides the node network topology,such as user portrait information in the social network and user blog information in the microblog network.in link predication task,the node attribute information are not fully considered,and related research is less.Therefore,in view of the above problems,this paper takes "how to fully and effectively utilize network topology information and node attribute information for link prediction" as the main research content.The specific research work is as follows:A method of link predication based on network embedding is proposed,which named NELP(Network Embedding Based Link Predication).The network embedding technology is applied to the research of link prediction problem.For the problem that the structural similarity index is difficult to be universal,the topology information of the network is embedded into the low-dimensional and dense vectors by network embedding algorithm.This unsupervised representation learning method captures network topology information more completely.The node's latent feature vector is further used to generate the connected structural feature vector for link prediction.The experimental results verify that the above method makes full use of the network topology information and improves the accuracy of the link prediction task.A method of fusion topology and node attributes is proposed,which help to combine fusion topology and node attributes to conduct link predication.The algorithm performs unified coding on the node attributes in the network to obtain the node attribute feature vector,and then passes it to NEAEF-LP(Network Embedding with Attribute Early Fusion Link Predication)and NEADF-LP(Network embedding with attribute deep fusion Link Predication)methods to conduct Link prediction.NEAEF-LP use the early-fusion method to concat network structure features and node attribute features;NEADF-LP input weighted structural features and attribute feature into the deep neural network for nonlinear-fusion and outputs link prediction scores.The experimental results verify that the addition of attribute information improves the accuracy of the link prediction algorithm,and the NEADF-LP method which using neural network for deep feature fusion achieves higher prediction accuracy on several attributed nework.In this paper,the proposed link prediction framework combine network topology and node attribute information for link predication task.The effectiveness of the algorithm is verified by experiments.The research work is meaningful.