Abstract:

Mining the collaboration patterns on social networks has been studied extensively in recent years. Collaboration patterns are manners of how individuals collaborate with each other, and such patterns can be represented by graph substructures. In some existing studies, including frequent subgraph pattern mining, only the structure pattern is  considered, and a minimum support should be given for controlling the scale of results. In some cases, interesting patterns could not be frequent, and exactly matching between patterns and communities is also unnecessary. We consider the social positions of community members, and give a pattern specification on weighted graphs. We propose a similaritybased pattern matching measure, and our goal is to enumerate all the representative collaboration patterns based on that. We design a distancebased clustering method to retrieve collaboration patterns, and we verify our algorithms on a large real data set.

Key words: collaboration pattern；social network；clustering；similarity measurement；weighted graph