Research And Application Of Collaborative Filtering Algorithm Based On User’s Search Content For Movie Recommendation System

With continuous development of internet technology, e-commerce、social network and vertical portals need technology to do some personalized recommendation for users, so that operators can earn more revenues and improve user experience. These technologies are personalized recommendations. Following the growth of diversity of recommended content, the increase demand of movie recommendation is needed. However, researches on movie personalized recommendation based on collaborative filtering are rare and there are still many problems to solve on movie personalized recommendations based on collaborative filtering method. This thesis aimed at the deficiencies of movie recommendation based on collaborative filtering, such as sparse feature, low precision of user similarity and so on, to propose some solution to improve precision of movie recommendation.First of all, the development process, framework and category of collaborative filtering technology were described in detail. Then the problems of collaborative filtering such as sparse feature, algorithm scalability, user security, cold boot and real time requirement were analyzed in detail. Secondly, according to the problem of user security, an algorithm for feature-weight valuation based on keyword-movie categories dictionary and user search records was proposed. User search vector was firstly converted to user-movie type vector, and then the value which represents the target user’s preference for particular movie genre was assigned to the original user feature matrix as the initial score for the item that never be scored. Thirdly, according to the low precision of user similarity computing of collaborative filtering, an algorithm which computes user similarity based on Chinese thesaurus extended edition and user search record was proposed. With computing the correlation between sememe, words and search vectors based on Chinese thesaurus extended edition, the User similarity was obtained, and a method that computes user similarity combining user score and search vector was finally completed.Eventually, to verify the improved method, a movie personalized recommendation system based on B/S was designed and three comparative experiments were carried out on this system by using Movie Lens as test set. Experimental results demonstrated that the two improved algorithms for collaborative filtering proposed by this thesis can actually improve precision and user satisfaction of the recommendation system. Simultaneously these results also demonstrated that the researches of this paper had practical significance.