A method of personalized recommendation based on multi-label propagation for overlapping community detection

Collaborative filtering among the methods of personalized recommendation was based on the entire user network that produced large amounts of operational data, and then led to the problem which recommendation efficiency was relatively low. To solve this problem, bipartite network composed by users and items in recommended system was mapped into synthetic user network, and then we detected overlapping community of synthetic user network. Multi-label propagation algorithm for overlapping community detection proposed by this paper was the extension of LPA. For detecting overlapping community structure MLPAO let each node with multiple labels and made updated labels of each node store in the memory of the node, and all the labels in the memory played a role on label updating of its neighbors. We selected asynchronous updating strategy, and utilized node preference to weaken the influence brought by the randomness of updating orders for enhancing the robustness of MLPAO. When the algorithm stopped, overlapping community structure of synthetic user network could be obtained from post-processing based on labels. In the overlapping community structures we recommended the target user items with collaborative filtering based on Pearson similarity. At last we compared recommended accuracy and recommended efficiency of the two methods with the MovieLens data set for the testing data. The results show that recommended efficiency of collaborative filtering based on community detection is essentially enhanced where recommended accuracy on line is almost unchanged.