Modeling Opportunistic Social Networks with Decayed Aggregation Graph

The emerging mobile applications including contact-based forwarding and online worm containment have put a heavy burden on opportunistic social networks (OSNs), the first goal is to develop an effective model that can reveal the hidden, important social features underlying the OSNs. This problem is especially challenging because of the time-varying network topology. Traditional time expanded graph caches each snapshot of networks, resulting in low computation efficiency and high storage overhead. By aggregating past contact events between two nodes to a simple boolean indicator, the binary graph model can alleviate this issue. However, it only provides a coarse-grained level of identifying the relationship between nodes. It neglects the differences in contact events. Intuitively, recent contact events are generally more important than old ones, and in computing an aggregation, we should assign bigger weights to them. In addition, since each contact has its own duration, we should take this factor into account as well. Motivated by these observations, we propose DAG, a decayed aggregation graph for modeling OSNs at a fine-grained level. By implementing DAG in different real scenarios, we show that DAG is efficient in characterizing the relationship between nodes and in improving the performance of mobile applications. We simultaneously prove that DAG achieves approximate space complexity as the binary graph.