The Study Of Video Recommendation Algorithm Based On Topic Model Under The Smart TV Side

With the development of Internet and terminal, Smart TV has replaced the traditional TV as the main way to watch videos. The Smart TV has richer channels to obtain videos than the traditional TV. Thus, the amount of videos can be viewed has been a sizeable growth compared with the past, but it brings the "information overload" for the users to watch. How to find the video users need from the massive video resource has become a major issue for the smart TV users currently. Based on this issue, it is the most im-portant to study the video recommendation algorithms for the smart TV users.Topic model can mine the semantic information of the text. Applying topic model to the video recommendation is proposed, which can mine the potential interest of users and improve the issues of the traditional algorithms to recommend at the smart TV side. Two ways to construct interest topic model of users are proposed. One constructs inter-est model of users dynamically based on the LDA model. Another constructs interest model of users based on Unified Topic Model (UTM-LDA) which base on the perspec-tive of users. UTM-LDA has improved the LDA model, and when constructing the in-terest model of users can unify different behavior data of users.Recommending videos for users base on their interest model. User’s interest may change over time, especially for the smart TV shared by multiple users, which result in diverse and different interest of the user. Thus a dynamic recommendation algorithm based on dynamic user interest model is proposed. But the Smart TV side just has im-plicit behavior data of users, based on which a recommendation algorithm unifying us-ers’more implicit behavior data is also proposed. The experimental result indicate that two recommendation algorithms proposed are better than the basic algorithm based on LDA, and especially for the recommendation algorithm based on UTM-LDA is the best.