Towards simultaneously exploiting structure and outcomes in interaction networks for node ranking

In this paper, we present algorithms for ranking nodes in interaction networks. Informally, they capture the patterns of historical interaction among the nodes and the associated outcomes. There exists a cardinal ranking over the set of outcomes, characterizing the order of preference. We argue that ranking of nodes should be influenced by both structural properties of the networks and the outcome/value created by the interactions. The former aspect is well studied in social network analysis and is accounted for, in various measures like centrality, reputation, influence etc. However, the latter aspect is largely unexplored. Our proposed algorithms simultaneously take into account both structural properties as well as the outcomes to assign ranks for the nodes. We develop a novel eigenvector-like computation that exploits the structural influences, importance of value creation, and any exogenous information available to the ranking system. We report experimental results on the IMDB dataset.