A Method for Latent-Friendship Recommendation Based on Community Detection in Social Network

Abstract

The paper studies a method for recommendation based on community partition applying for user in social network. Firstly, the largest connected component in friend-relationship complex network are taken as the logic unit, and divide up the largest connected component into non-intersect kernel sub-network, the kernel sub-network based on The maximum complete sub-graph which has the mathematics foundation and convenient for the promotion of this algorithm. Secondly, create labels for each node outside the kernel relationship after the label spreading based on the kernel sub-network. In addition, calculate the weights of labels at nodes for eliminate the labels which weights are too small by self-adaptive threshold. In the end, recommending each other between the latent friend-relationship after finishing the community partition according to the label. The paper designs the related simulations and experiences in friend-relationship complex network at Scholat.com, to show feasibility, stability and robustness of Recommendation Method based on Community Partition, in the considerable efficiency. Further, we calculated precious, recall and F1 score according to the feedbacks from users to show the recommendation is accuracy.