Research And Implementation Of Wen Recommendation System Based On Hybird Recommended

With the development of information technology and the Internet, peoplegradually enter from information weary toward the era of information overload. Itbecomes increasingly difficult for user to search for the information they areinterested in the mass of information.Personalized Recommendation Service as a aneffective means to solve to information overload came into being,it usesrecommendation algorithm automatically find the content and services that meet theuser’s personal interests in the mass of information quickly.Currently, collaborative filtering or content-based recommendation is generallyused in personalized recommendation system. However, Collaborative filtering havecold start problems, with the increase in the number of projects and users, user-itemmatrix and become sparse, which will affect the accuracy of the recommendationsystem. Content-based recommendation algorithm needs to analyze the contents ofitems，which can not take fully account of the user’s interest. For collaborativefiltering and content-based filtering algorithm respective defects, hybridrecommendation algorithm can learn from each other, thus become an importantdirection of the recommendation algorithm research.In this work, some improved methods is proposed to i deal with above questionto prove the recommendation accuracy, the main work is as follows:1Traditional collaborative filtering just uses user-item rating matrix torecommend, recommendation accuracy is not high due to the data sparsenessproblem.we present a recommendation algorithm called ASVD,which add contentinformation to maximize the use of known information in the same model usingmatrix decomposition technique,which Alternately merge and optimized item-latenttopic matrix which is from the process of using singular value decompositionalgorithm to decompose the project-keywords matrix based on item content anduser_item rating matrix to eliminate the noise to improve the accuracy of therecommendation.2User similarity calculated in general collaborative filtering suffers from datasparsity and poor prediction problem. we propose use user’s feature vector tocalculate the similarity to alleviate the data sparsity, thereby enhancing the system’srecommendation precise degrees. Because calculating similarity in the whole domain in common collaborative filtering causes large amount of computation, this algorithmthrough the K-means algorithm to cluster users according to the user characteristicvector to find user’s nearest neighbor set, thereby reducing the amount ofcomputation..3Finally, build a web personalized recommendation service system, combinedwith the algorithms above to recommend.