Recommendation via Query Centered Random Walk on K-Partite Graph

This paper presents a recommendation algorithm that performs a query dependent random walk on a k-partite graph constructed from the various features relevant to the recommendation task. Given the massive size of a k-partite graph, executing the query centered random walk in an online fashion is computationally infeasible. To overcome this challenge, we propose to apply multi-way clustering on the k-partite graph to reduce the dimensionality of the problem. Random walk is then performed on the subgraph induced by the clusters. Experimental results on three real data sets demonstrate both the effectiveness and efficiency of the proposed algorithm.

• Type: Conference
• Year: 2007
• Where: ICDM
• Authors: Haibin Cheng, Pang-Ning Tan, Jon Sticklen, William F. Punch
• Keywords: Data Mining, Dependent Random Walk, ICDM 2007, K-partite Graph, Random Walk