Research And Design Of Hybrid Video Recommender System Based On Social Network

For big data sets in UGC video website,how to find the videos that users want to watch?As usually,with the help of search engine or traditional recommendation.But there are some problems:the keyword text model is not meet the user’s interest,weak relate recommendation;a large number of disorder ranking search results;traditional recommendation is not show the personally interest.For no clear target user,how to find the video that they trully want? Academics and Internet companies have developed a variety of personalized recommendation system to handle this problem.The recommendation system get the user’s website activities,extract the appropriate behavior, build interest model, and predict user’s future interest and make a related recommendation. With the popularity of social networking, social network-based personalized recommendation system, there are still many of the key problems to be solved.To this situation, we research the hot recommendation algorithms in this paper,and construct algorithm to solve some problem like data sparsity and cold start problem.New recommendation algorithm is proposed in design the social network graph model, that is used to improve the video recommendation quality.This article has mainly done the following work.1. We analysis and summarizes the common personalized recommendation algorithm and similarity algorithm.The study concluded the composition and evaluation of personalized recommendation system based on user interest, domestic.We summarized the personalized recommended research status.2. An improved feedback model is proposed in design the user interest model in this personalized recommendation system.With traditional psychology,we described the transforming relationship between users’ long-term interest and short-term interest.And evolutionary pseudo-code description of the model.3. The graph model and Improved VideoRank algorithm base on social networks is proposed.With that algorithm,video recommended result is much better.We elaborated the key issues in extraction information,ranking videos,feedback verification.4. Matrix decomposition is improved to be used in item-based collaborative filtering recommendation algorithms.effective upgrade the recommended diversity and novelty.The subject model improved recommended results.With linear fusion the recommended queue,that adaptd the system application.5. Detection model is improved to apply the personalized recommendation system.We analysised the social networks attacks,indicated a number of effective advice and experience in engineering design.At last,in online environment, we carried out simulation tests.This solution is good at improve your videos clickthrough rate and returning rate,reduce the over-fitting of the recommended results.For large-scale social network data sets,it work well.