Mari Saito

Abstract:
In the automatic metadata annotation we propose effective in content recommendation matched to user preference, content is clustered based on user evaluation patterns for sampled content, not on the content itself. Non-sampled content is also annotated into proper clusters after metadata-based prediction. Verification of our approach using wine recommendation showed that our metadata was the most effective among many sets of metadata.

Keywords: recommendation, personalization, preference learning, content-based filtering, collaborative filtering