Reseach On Video Recommendation Technology Based On Data Mining

Video as a vivid information carrier is becoming the best choice for people, so video related service is more and more popular. The explosive growth of video related service and its contents makes users feel like a fish out of water. While, the rapid growth of users also makes the video service operators face enormous challenges and opportunities. To resolve these contradictions, video recommendation technology comes into being, which can help users quickly find their favorite videos and help operators promote their video resources to users. Nowadays, the video recommendation technology has achieved great success on some websites, such as MovieLens, YouTobe, Baidu,and so on.This thesis will study the meaning and the current developing situation of the video recommendation technology. And then carry out the technology research of video recommendation based on user-based collaborative filtering, latent factor model and clustering, to solve the problems of data sparsity and efficiency in the mainstream collaborative filtering.The key problem of user-based collaborative filtering is to calculate the interest similarity between users. In the traditional Top-N, user-based collaborative filtering only consider the implicit behavior data of users, without the explicit scores and user features. This thesis will study the improvement of similarity measure between users.Technology of video recommendation based on latent factor model, learn the relationship between users’ interests and videos through the latent features, then generate recommendations. This thesis will combine the neighborhood collaborative filtering with latent factor model to further improve the quality of recommendation.In addition, the numerous users and videos will take much more time to recommend videos. User clustering can improve the efficiency. However, in the conventional clustering-based collaborative filtering, clustering is done on the scoring matrix.The calculation of this method is still great. This thesis will simplify the representation of users’ interests through video features to improve the efficiency of clustering.In summary, the main work of this thesis are the following three aspects:① After analyzing user-based collaborative filtering, consider users’ explicit scores on the videos and user features to calculate users’ similarities. A detailed simulation analysis is carried out.② Combined latent factor model with neighborhood informations of videos and users, the precision of recommendations is improved effectively.③ Use video features and users’ scores for clustering users,and then do video recommendation in every cluster, which has greatly improved the efficiency of recommendations.