A Multi-information Embedding Link Prediction Approach with Collective Attention Flow Network

Link prediction and community detection reveal the basic mechanism and evolution law of the network from different perspectives, and the relationship between community members can provide valuable information for link prediction. Most link prediction methods are based on local structural features and lack the application of community topology information. To deal with this challenge, we propose a link prediction method NCELP with collective attention flow network. Firstly, we apply the Louvain algorithm to give each node a community label as an explicit feature. Secondly, using local structural features and community topology information, learned node and community embedding, which serve as implicit features. Finally, combined with the graph structure features, the link prediction problem is transformed into a binary classification problem and realized by the existence probability of the edge. We validated NCELP using behavior data from China Internet Network Information Center with more than 30,000 online users and three public datasets. Experimental results verify that NCELP not only outperforms the state-of-the-art methods on real-world datasets but also improves its AUC value by at least 9.83% and its AP value by at least 3.39%.