Abstract: In view of the traditional community detection algorithms being mainly applied to single relational networks, ignoring the interaction of relationship in the multi-relational networks, being unable to distinguish the importance of each relation for community detection, a novel algorithm called InteractRank was proposed. Based on the node and the relation of ranking model, the algorithm could transform multi-relational network into single relational network. Combined the PageRank algorithm and the random walk model, the algorithm considered the connection within groups and between groups in multi-relational networks. After transforming into single relational networks, spectral clustering algorithm was adopted to detect community. Through the simulation experiments on the standard UCI dataset, InteractRank indicates to be effective to community detection in multi-relational networks.

Key words: multi-relational networks, community detection, PageRank, random walk model, spectral clustering