The Research Of Social TV Recommendation Based On User Behaviours

With the rapid development of mobile Internet and social networks, more and more users communicate and interact with each other through social network, making social network gradually become the dominant force of creating,spreading and obtaining information in the Internet.As a traditional media, TV has never stopped its development, actively looking for the integration with new media. And social TV emerges as the times require.Social TV has changed the information dissemination way of traditional TV media. Users can discuss the TV programs through the social network at any time and any place, making TV become an interactive and popular communication platform. In the Social TV scenario, users publish and get TV programs information in the social platform. Then, by a large number of users'forward, the information spreads quickly, producing vast amounts of social data which can be used to generate personalized recommendation. Under the above background, the thesis studies the social TV recommendation based on user behavior.A recommendation model of implicit feedback based on social behavior and tag is established by analyzing user feedback. Focused on the rich user reviews of social TV, the model adopts sentiment analysis to map text into virtual ratings, and implements personalized recommendation with Probabilistic Matrix Factorization (PMF) model. At the same time, taking the cold start problem into consideration, the thesis introduces the user tags, and proposes a Weighted SemanticTag Similarity Method (WSTSM) to compute the semantic similarity of tags which is regarded as user similarity. And it is added into the regularization of PMF model.However,the virtual ratings got by above algorithm ignore the specific interest of user to program. Thus, the thesis further improves the algorithm by introducing the topic model before executing sentiment analysis. This topic model can extract the user interest on the feature of program to refine the user interest model, improving the recommendation accuracy.Finally, based on the above model of social TV, a social TV recommendation system based on user behavioris designed and developped,realizingthe personalized recommendation in social TV scenario.