Contextual Recommendation with Path Constrained Random Walks

Recommendation has become an increasingly important research area because of the abundance of online documents and shopping opportunities. The readily available context information for these tasks posts challenges to recommendation systems to deal with heterogeneous and inter-connected problem spaces. In this study, we use a graph representation for publication databases with rich meta-data. With this representation, a Path Constrained Random Walk (PCRW) model is trained to discover effective recommendation strategies represented as edge paths on the graph. Experiments on both citation-based and history-based reading recommendation tasks show that by leveraging rich context information the PCRW based approach outperforms random walk with restart based approachs as well as traditional content-based and collaborative filtering approaches. An online recommendation system for Saccharomyces Genome Database is available at http://hops.ml.cmu.edu:8080/strutsnies.