Research On Collaborative Filtering Recommendation Algorithm Based On Time Series Analysis Of User Rating

Along with the advent of the era of web2.0,interactive network is increasing dayby day, now users are no longer just passively receiving information, but activelyinvolved in writing Internet content. In recent years, information content raisesexponentially, the information contribution by different people, all over the world,which has increased the difficulty of the users to search information. In the vastamounts of information world, how to filter fast and find the information they areinterested in have become the focus in the current theory and practice. A lot ofresearch use recommendation system to solve this problem. The existingrecommendation system that has been applied, need to recommend too much projects,and the number of users is increasingly large, it causes that the system must face thehuge data and extremely sparse, leading to low efficiency and precision of projectdelivery process. Therefore, how to enhance the r quality and efficiency ofrecommendation system to improve the quality of service, become a problem to besolved.The most important part of the recommendation system is the algorithm. Thealgorithm to some extent can decide the efficiency and performance of the system, sothe choice of recommendation algorithm determines the quality of the system.Traditional Collaborative Filtering (the following will use shorthand CF)recommendation algorithm is limited to users-project evaluation matrix, almost noscore information with the specific content of the properties. In the extreme conditionthat data are sparse, traditional recommendation algorithm in personalizedrecommendation seriously insufficient, data sparse sex of neighbor search accuracy isnot high, which seriously affected the effect of recommendation system.In order to solve this problem, this paper conducted the following research:First of all, by calculating statistics, we found that the deficiency of thetraditional methods for searching the neighbors: the extremely sparse data errorcaused by the traditional algorithm. This article puts forward the neighbor calculationmethod based on similar user behavior. Second, this article improved the problem that traditional CF algorithmsrecommend accuracy is not high under the extremely sparse data condition. Thispaper put forward the optimization algorithm that combine the traditional CFalgorithm with the Controversial similarity algorithm.Finally, through the experiment, this paper found that the optimization is obviousbetween the recommendation algorithms that combine dispute based on similaritywith similar user ratings behavior and traditional CF recommendation algorithm basedon users.