A Top-k Similarity Node Search Method Based on Convolutional Neural Network in Complex Network

Abstract

AbstractIn complex network, it is challenging to find similar nodes for a given node. Existing methods focus on evaluating the structural similarity between nodes but ignore their textual similarity, which makes returned nodes are not accurate enough. In order to search top-k nodes with similar text and structure for a given query node in complex network, we propose a similar node search algorithm based on convolutional neural network (LRE-CNN). For a weighted complex network, we first construct the nearest neighbor network model based on degree and weight for each node. On this basis, relative entropy and joint tightness are used to evaluate the comprehensive structural similarity between nodes. We get the candidate nodes with the highest structural similarity to the query node. Next, convolutional neural network is used to extract and compare potential text features between query node and candidate nodes. Finally, top-k similar nodes are returned for the given query node. Through experiments on complex networks of different scales and comparison with existing node similarity measurement methods, the results demonstrate that LRE-CNN has high retrieval accuracy and high execution efficiency. It can be effectively applied to top-k search of similar nodes in large-scale complex network environment.KeywordsNode similarityTop-k queryConvolutional neural network