The Improvement And Application Of CF Algorithm Based On Weighted Bipartite Graph And K-Medoids

With the advent of the Internet era, people have been more and more familiar with the concept of "big data". All aspects of the issues in Recommendation system, such as sparse data, drift of users'interests and poor efficiency of recommendation, etc., have emerged with the soar of user scale. Aiming at these issues, this dissertation has studied on collaborative filtering recommendation algorithm. The main research works is presented as below:(1) Discuss the principle of collaborative filtering algorithm (CF) and present a new collaborative filtering algorithm which is based on weighted bipartite graph algorithm (WBG-CF). In this algorithm, user behavior time data is used to be weighted with user-project matrix and then the matrix is filled by using bipartite graph algorithm, which is followed by finally conducting collaborative filtering recommendation. Sparse data and drift of users' interests in collaborative filtering recommendation algorithm are solved by this approach. And the test show that the improved WBG-CF algorithm can reflect drift of users'interests, whereas the accuracy rate of CF algorithm is also enhanced.(2) Aiming at poor efficiency of recommendation on WBG-CF algorithm that is due to the soar of user quantity, a new WBG-CF algorithm that is based on K-Medoids clustering has been proposed. In this algorithm, after filling the matrix by using weighted bipartite graph algorithm, a cluster to users is conducted and then k user clusters is recommended by using collaborative filtering recommendation. The results of experimental show that KWBG-CF has certain advantages in processing efficiency on medium data set.(3) Develop an application by using improved algorithm in order to perform visual interface display for the results of recommendation algorithm, further verifying the feasible of algorithm.