Vertex Collocation Profiles Subgraph Counting for Link Analysis and Prediction.pdf

WWW 2012 – Session: Social Networks April 16–20, 2012, Lyon, France Vertex Collocation Profiles: Subgraph Counting for Link Analysis and Prediction Ryan N. Lichtenwalter Nitesh V. Chawla University of Notre Dame University of Notre Dame Notre Dame, Indiana, USA Notre Dame, Indiana, USA rlichten@ nchawla@ ABSTRACT and target vertices. This idea is a generalization and exten- We introduce the concept of a vertex collocation profile sion of the triangle counting approach for multi-relational (VCP) for the purpose of topological link analysis and pre- prediction in [6]. It also draws on concepts from literature diction. VCPs provide nearly complete information about on graphlets as introduced in [19] and to a lesser degree from the surrounding local structure of embedded vertex pairs. motif analysis as discussed in [17]. The VCP approach offers a new tool for domain experts Many existing link prediction models compress a selection to understand the underlying growth mechanisms in their of simple information in theoretically or empirically guided networks and to analyze link formation mechanisms in the ways. By contrast the VCP approach preserves as much appropriate sociological, biological, physical, or other con- topological information as possible about the embedding of text. Th