Relevance Modeling of Linked Open Data and Users' Transaction Histories for Recommendation

概要

論文の詳細を見る
We propose using Linked Open Data (LOD) to inform a topic model for recommending both well-watched old movies and unwatched new ones. For old movies, users' transaction histories can effectively inform recommendation, while for new movies, the absence of such data makes content-descriptive information necessary. In our algorithm, we address the issue of finding movie metadata by automatically drawing it from LOD. Then, our metadata-relevance-sensitive model uses the relationship between the transaction history and metadata to make recommendations. We compare our method's recommendation accuracy against conventional methods found in the literature, and find that only our method is able to successfully use LOD to recommend both well-watched and unwatched movies.
2012-03-01