Collaborative Filtering Algorithm Based On Weighted User Influence

Today, with the rapid development of Internet resources the network increases rapidly, which makes it easy to get access to information and resources,but with the media resources getting more and more, the growth’s rate of project resources greatly exceeds the growth of users’ number, so it’s difficult for users to accurately find the information they want from massive resources, this is the problem of information overload. As one of the main methods to solve this problem, recommendation systems can help users to lock their interesting informations from resources faster and more accurately. Recommendation system technology studies users’ interest categories and behavioral models, and then recommends resources to users which meet their interests and habits.As the most successful promoted recommendation technology in generous recommendation systems in 21 st century, collaborative filtering technology has been great valued by the industry experts.As the name suggests, collaborative filtering means that users can work together,by means of interact with website constantly, expand their own items list of recommendations which they are interested in, so the list meets their needs increasingly. In general there are several major technical problems existing in traditional collaborative filtering algorithms, among these problems the most important several issues include the data sparseness problem, cold start problem, and scalability issues. This thesis analysis the application and research status of recommendation systems, classifies and summarizes several of the main methods. This thesis studies various collaborative filtering algorithms deeply, inspired by the complex network methods based on collaborative filtering recommendation algorithm using graph model, we propose a collaborative filtering recommendation algorithm based on users’ influence weight, the algorithm takes advantage of the users’ implicit datas such as purchase, browse, click to construct a users- projects network, this method can describe the structural correlation of the users-items network systematically, at the same time introduces the forgetting curve of time weighted function based on the time context, gives a method to measure the users’ influence,finally, combined with the similarity based on explicit data including users’ ratings to make the final weighted recommendation. Compared to the collaborative filtering based on the theory of heat conduction,collaborative filtering algorithm based on diffusion theory and the collaborative filtering algorithm integrated singularity and mass diffusion,the proposed method can alleviate the sparsity of project scores, while improves the algorithm’s performance both in diversity and accuracy,as a result of the characteristics of the complex network method, the calculation method is easy and intuitive to understand, easier to calculate a large number of users and items add, and has a good scalability.Finally, this article details all aspects of the final recommendation from data modeling to the pseudo-code, and codes in JAVA language for this algorithm which is based on real experimental data sets,experimental evaluations on multiple comparison with other algorithms show the feasibility and accuracy of the algorithm.