Dynamic Community Detection Algorithm Based on Automatic Parameter Adjustment

Community detection is widely used in social network analysis. It clusters densely connected vertices into communities. As social networks get larger, scalable algorithms are drawing more attention. Among those methods, the algorithm named Attractor is quite outstanding both in terms of accuracy and scalability. However, it is highly dependent on the parameter, which is abstract for users. The improper parameter value can bring about some problems. There can be a huge community (monster) sometimes; other time the communities are generally too small (fragments). The existing fragments also need eliminating. Such phenomenon greatly deteriorates the performance of Attractor. We modify the algorithm and propose mAttractor, which adjusts the parameter automatically. We introduce two constraints to limit monsters and fragments and to narrow the parameter range. An optional parameter is also introduced. The proposed algorithm can choose to satisfy or ignore the optional parameter by judging whether it is reasonable. Our algorithm also eliminates the existing fragments. A delicate pruning is designed for fast determination. Experiments show that our mAttractor outperforms Attractor by 2%–270%.