Lasso—Based Tag Expansion and Tag—Boosted Collaborative Filtering

摘要

With the increasing popularity of the social tagging systems, tags can be effectively utilized to enhance Collaborative Filtering (CF) algorithms. Tags not only reflect users' preference, but also are a cue to describe the semantics of items. This paper formulates the problem of collaborative filtering as random walks over the user-item-tag tripartite graph. In order to alleviate the sparsity of tags, a lasso logistic regression model is conducted to accomplish tag expansion, i.e., adding relevant tags and removing irrelevant tags for each item. Experimental results on MovieLens dataset demonstrate the superiority of the proposed algorithms over several existing CF algorithms in terms of ranking performance measure Fl and Macro DOA.