Research On Collaborative Filtering Personalized Recommendation Algorithm Based On Clustering

With the rapid development of information technology and the Internet, information resources have gone into the level of index growth.In the face of serious overload of information, the user is difficult to quickly locate the useful information resources for them, which cost a lot of time to search the content they want. This makes the recommendation system came into being, it can provide personalized recommendation service based on the preferences of different users. Currently, in many personalized recommendation technologies, collaborative filtering algorithm demonstrates its unique advantages, so it is widely used, particularly in the area of electronic commerce it has achieved considerable success, but traditional collaborative filtering algorithms have data sparse, cold start, real-time and other problems, if they can be effectively overcome, not only to improve customers’ satisfaction, but also increase sales profits.In this paper, on the basis of the recommendation system, with recommendation technology as the main line, make research on the personalized recommendation system, personalized recommendation algorithm and related technologies, and make comparative analysis; Then, for the problems of the better applied collaborative filtering algorithm at present, introduce project properties and user characteristics to build the project properties matrix and user characteristics matrix, propose calculation method based on project-related similarity and user-related similarity; addition, make related research on clustering technology, use Kruskal algorithm to improve the traditional K-means clustering, so it could automatically determine the initial cluster centers; Finally, based on user and project direction clustering, carry out relevant research, propose to combine the initial forecasts based on project properties clustering with the final prediction based on user characteristics clustering to make a recommendation. The main research work has been done in the following aspects:First, in view of the traditional collaborative filtering algorithm relies too much on user-project evaluation matrix will face serious data problems about data sparseness and cold start, in this paper, make full use of the user characteristic information and the project properties information to avoid the phenomenon of "similar but not identical" as well as to overcome the problem of new projects and new users, put forward respectively correlation similarity computing method based on the project properties and based on user characteristics;Second, the traditional K-means clustering were studied, in view of the problem of traditional K-means clustering be sensitive to the initial clustering center for random selection, put forward a kind of improved clustering algorithm which can automatically generate relatively evenly distributed K initial clustering centers;Third, to effectively reduce the neighbor query space in collaborative filtering algorithm, using the improved clustering algorithm to cluster respectively in the project properties matrix and user characteristic matrix at the same time, significantly reduce dimension calculation, improve the efficiency of recommendation;Fourth, in view of the problem of user rating data sparseness, combine with the results of neighbor searching based on project clustering and original rating matrix, make the initial forecast based on project to fill the original rating matrix. Among them, new users or new projects, use correlation similarity base on project attributes or user characteristics instead of the traditional rating similarity to search neighbors, overcoming the problem of cold start; Finally, combine with the results of neighbor searching based on user clustering and rating matrix filled well to make the final score prediction based on user, causing the precision of recommendation higher.Fifth, in order to verify the effectiveness of the improved algorithm is proposed in this paper, the improved collaborative filtering algorithm has been analyzed and carried on the contrast experiment with some traditional collaborative filtering algorithms on the selected MovieLens data set respectively. Experimental results show that this method can effectively solve the problem of the data sparseness, cold start and scalability problems, making the final recommendation quality is better than other traditional recommendation algorithm.