Retrieval of Relevant and Non-redundant Nodes ∗

We discuss the problem of discovering interesting nodes in networks. We adapt a generic model to choosing relevant and non-redundant pieces of information in networks and probabilistic relations. In the model we assume that one or more query nodes have been given, and the problem is to identify other nodes that are relevant with respect to the query nodes but non-redundant with respect to each other. Also, negative query nodes can be specified. This is in contrast with mainstream graph mining, where one typically looks for frequent patterns, not for interesting individuals. We consider two instances of the model: one where node proximity (and relevance) is measured by the shortest path, and one where the graph is probabilistic or uncertain and proximity reflects the probability that the nodes are connected. The generic model also has simple parameterization to allow for different behaviors with respect to query nodes. We compare different similarity measures and empirically evaluate two algorithms on different applications: social networks and biomedical networks. The results indicate that the model and methods are useful in finding a relevant and non-redundant set of nodes.