Research Of Collaborative Filtering Recommendation Algorithms Based On Satisfaction Similarity

In recent years, with the Internet-based platform, with the growing popularity of electronic commerce sites, on-line payment is becoming indispensable in people’s life, as well as on-line shopping. How to make the recommendation system recommend items to users which they are interested in more accurately and more quickly has become the key point to build a solid "bridge". It is also the purpose of studying about this kind of algorithms for many researchers. This paper just studies around this issue deeply.Firstly, this paper discusses a variety of recommendation techniques used in the personalized recommendation systems, especially the collaborative filtering technique, and makes its principle and classification understood fully. In addition, this paper introduces several classical algorithms and their process at length, and researches and analyses their advantages and disadvantages, and then gives some ideas for improvement.Secondly, according to research the collaborative filtering algorithms, this paper finds some problems of the traditional algorithm and then proposes an improved collaborative filtering algorithm based on satisfaction-similarity combined with every user’s personality. So that this paper has combined every user’s characteristic with the similarity between users, and then grasps the information of user’s interests efficiently through user’s history data, and takes the influence of the number of common items on the similarity between users into account to make the prediction of users’ rating more accurately.Thirdly, to solve the problem of data sparsity faced by collaborative filtering recommendation algorithms, this paper proposes an improved algorithm combined the satisfaction-similarity with Slope One, which makes up the problem of Slope One ignoring the similarity between users.Finally, this paper works on a verification study using MATLAB program, gets the user’s predict rating scores by using the methods proposed, and then analyzes the results of the two evaluation standards, MAE and RMSE. At the same time, this paper lets the method proposed compare with other algorithms and then gets the result to verify the feasibility and effectiveness of the optimized algorithm.