Research On Link Prediction Algorithm Based On Network Features Fusion

Link prediction plays a practical application value in different fields such as virus propagation prediction,traffic flow prediction,friend recommendation,and fraud detection.It is a meaningful work to study link prediction algorithms with high robustness and high accuracy.Most of the traditional link prediction algorithms are proposed based on the similarity between nodes.The applicable network is limited,and it is difficult to adapt to networks with different characteristics in different fields.In order to improve the prediction accuracy and the robustness of the prediction algorithm,the paper uses network embedding technology to learn the low-dimensional dense representation of the network,and combines the characteristics of the network,using community structure information,community relations and common neighbor community information,The link prediction algorithm of merging network features is studied.The main work of the paper is as follows:(1)Based on the analysis of representative link prediction algorithms in the past ten years,the paper reproduces a total of 17 representative link prediction algorithms such as the Salton index,through experimental comparison,it is found that the above methods show large differences in performance when predicting different networks,and then analyze the factors that affect the prediction effect.(2)Using network embedding representation,the paper proposed a Link prediction algorithm with network features fusion based on network embedding(LP-NFNE).First,the network embedding technology node2 vec is used to obtain a vectorized representation of the node topological position features,and it is fused with the network features to construct a feature vector describing the edge features,the classifier uses a random forest model for training,and then realizes the prediction of unknown edges.Finally,12 public networks such as USAir are used as the experimental network,and the experimental comparison with 17 representative link prediction algorithms verifies the performance and robustness of the algorithm in this paper.(3)A link prediction algorithm combining community relations and common neighbor community information weighting is proposed(CER).The paper studied the idea of using network feature information to improve the method of defining similarity based on common neighbors of nodes.The paper proposes a link prediction algorithm that combines community relations and common neighbor community information weighting,the algorithm combines the similarity of nodes,the similarity between the communities where the nodes are located,and the community information of the nodes' common neighbors to measure the link probability of two unknown nodes to improve the prediction effect.Finally,it compares with four groups of baselines such as Salton indicator on 9 networks with community structure such as USAir to verify the effectiveness of the algorithm.