Research On Video Recommendation Algorithms That Integrate Social Relationships And Content

As the rapid development of the Internet and information technology, people have slowly entered the era of information overload from the era lacking information. At the same time, the popularity of the social network has spread to our daily life. Watching and sharing videos are the common behaviors in social network. As for lots of videos in social network, traditional search technology can't meet the need of the users, especially for those users who have not definite requirement. The recommender system has emerged to solve these problem, which not only collects the information of users in social network, but also extract the habit of users' behaviors. And then the interest model is constructed to predict users' hobbies for making recommendation. Therefore, it is a popular research issue on combining recommender system with social network.This thesis mainly studies the algorithm of video recommendation with social relationship to solve the problem that the videos provided by the recommender system can't satisfy the demand of users for watching videos. In order to make the video recommender system perform much better, we improve the recommendation method by finding those friends trusted by target user and evaluating the quality of videos respectively.In the model of trusted friends computation, we take the users' similarity, users' interaction and active degree of user into account to find out the friends that target user can really rely on and make the recommended videos more reliable. With respect to the calculation of user similarity, the direct similarity and indirect similarity are distinguished. As for the user interaction, we have considered three factors as follows: @, retweet and comment. We also integrate with the tweets and fans of a user to derive the active degree of user. How to get the high quality videos viewed by trusted friend is another important research issue, because not all the videos from trust friend are suit for the target user. Our proposed model of quality evaluation for videos that combines the reception ratio with the reputation of videos.The reception ratio of a video represents the watching result made by users to this recommended video. And the reputation of a video evaluated by those users who have certain fans and viewed this video before. It can guarantee the quality of recommended videos through the quality evaluation model of videos. Moreover,our proposed video recommendation model can get more videos that users are really interested in and improve the experience for watching videos in social networkAt last, we conducte some experiments to demonstrate our algorithm by comparing with the algorithms of User-based CF and TBR-d algorithm. The results show that our proposed algorithm is superior to both User-based CF and TBR-d in terms of precision, recall and F1-measure. That is to say our proposed algorithm can recommend more videos that users are interested in.