Ontology-based personalized recommendation system

Abstract

Past decade has seen a prominent rise in the number of e-commerce applications in the World Wide Web. Designing recommendation algorithms for predicting user interests is quite challenging for such systems. Several recommendation frame- works have been proposed in research. However, when it comes to recommenda- tion of media-rich commodities, most of the algorithms designed so far, utilize the metadata associated with the digital products. Such systems may not generate correct recommendations if the metadata is insu cient or inaccurate. Our approach is motivated by the fact that by making use of a domain ontology and relating media content to domain concepts, it is possible to remove the semantic gap between high-level semantic concepts and low-level media features. This can be utilized to improve recommendation of media-rich commodities to the user, as such a recommendation is based on media content as well as metadata. In this work, we have proposed a video recommendation framework based on ontology. The multimedia ontology is represented in Multimedia Web Ontology language (MOWL), which supports a probabilistic reasoning scheme. We have also given a novel approach for personalizing the recommendations on-the-fly, by analyzing user preferences and modifying the recommendation model accordingly. We have experimented with a media-rich dataset consisting of English movie videos. Proposed system can add semi-automatic conceptual annotations to movie scenes as well as to full movies with the help of the ontology. This semantic metadata is also utilized while making recommendations to the user. The system can recommend not just full movies, but scenes from the movies based on user interest. We have illustrated the proof of concept by corroborating our system with anonymous users.The contentment score and recommendation accuracy obtained, has validated the efficiency of our approach.