Social Communication to Improve Group Recognition in Mobile Networks

In this paper, we consider the problem of recognizing groups of similar nodes within a mobile network. Nodes belonging to the same group have the same preferences, e.g., in terms of friendship with other nodes. However, due to limited communication range, each node has a partial view of the network and is aware only of its friendship relationships with the immediate neighbors. Depending on how large the communication range actually is, the problem can be in principle solved by clustering nodes by classifying the friendship relationships on a scale from -1 to +1 (sure not friendship to sure friendship, with unknown nodes classified as 0). However, if the communication range is small, this approach leads to large errors. We show how a simpler procedure based on the exchange of social information in a collaborative fashion is able to achieve better results than more expensive clustering algorithms. Also, these two approaches can be combined by clustering nodes after exchanging a given amount of social information, thus improving the overall results.