Research And Implement Of Recommender System Based On Collaborative Filtering Algorithm

In recent years, With the continuous development of computer technology andrapid popularization of internet, more and more informations can be exposed topeople, while the "information overload" problem has also cropped up. To solve the"information overload" issue, recommender system(RS) is needed. Compare with thetraditional type of search engine services, personality is the advantage of RS. Theresearch of RS is the urgent needs of the rapid expansion of information, and also hastheoretical value and practical value. Collaborative filtering(CF) technology is one ofthe hot topics in the research of RS, but it also has some shortages such as sparsityproblem and cold-start problem, at the same time there is still a considerable room forgrowth in the accuracy of the algorithm.Aiming at the above-mentioned problems, the main contents are RS and thesolution of sparsity problem, cold-start problem and accuracy improved. So we cansummarize the content and innovation point as follows:Firstly, the comprehensive study of RS concepts is the first step,include theresearch of related concept of RS, research of the various RS and their advantagesand disadvantages and research of CF algorithms which goes in deep.Secondly, based on the above-depth study on CF algorithm, we propose animproved CF algorithm through a combination of user’s and item’s attributes for thealgorithm of data sparseness and cold-start problems. This algorithm can takeadvantage of information of users and items of RS to make up for the lack of classicCF algorithm. Also unlike a simple combination, this algorithm can dynamicallyadjust the user’s and item’s attribute information in the score predicted proportionaccording to the characteristics of scoring data, making further improve in accuracy ofthe algorithm, and through experiments we verified this conclusion.Thirdly, through summarizing and analyzing the existing CF based RS, by takethe movie RS as an example, we design and realize a improved CF based RS and thentest the utility of the improved CF algorithm.