Publications

Abstract : Increasingly enlarged public distribution websites like Flicker and YouTube permit users to create, share, annotate and commentary Medias. The large-scale user-created metadata not simply facilitate users in distribution and organizing semantic attributes. Personalized search provides as one of such instances where the web search practice is improved by generating the returned list along with the modified user search intents. In this paper, we develop the social comments and propose a new structure concurrently considering the user and query relevance to learn to personalized image search. The essential basis is to embed the user preference and query-related search objective addicted to user-specific topic spaces. While the user’s original explanation is too sparse for subject modeling, we require enriching user’s clarification pool before user-specific topic spaces construction. The framework includes two components namely a Ranking-based Multi-correlation Tensor Factorization model(RMTF) is proposed to perform annotation prediction, which is believed as user's potential annotations for the images and further establish user-specific topic modeling to map the query relevance and user preference into the similar user-specific topic space. The performance evaluated based on the utilization of two resources for social activities.