Research On Personalized Recommendation Algorithm Based On Clustering

With the rapid development of Internet technology, the count of information on theInternet grow rapidly, the generation rate of the network information exceeds thereceived ability of people, Information overload seriously, mean while the quality of theinformation is uneven,which leads to the useful information is drowned in the ocean ofinformation. The search engine information retrieval technology needs people toprovide a clear search target, collaborative filtering take consider of the behalf of userthe personalized recommendation technology come into being, and is a effective resolveto solve Internet information overload problem.Traditional collaborative filtering algorithms predict the target user preferences foritems using the user and project ratings matrix to find the nearest neighbors, and it don’tuse of the user attribute information and project attribute information in e-commercesystems effectively, leading to poor accuracy rate, Finding the nearest neighbors of thetarget user is the key steps which needs to find the entire user space and leads to poorreal-time, besides, The algorithms is sensitive to the user rating matrix, the accuracydecline sharply in sparse high degree of systemClustering technology can effectively divided objects into groups, objects withhigh similarity is cluster in the same group, while objects with low similarity in theother group, the core steps collaborative filtering recommendation algorithm is to findthe similar neighbors of the target user, the application of clustering techniques whilerecommend can effectively improve the performance of recommendation.First, for real-time collaborative filtering recommendation, using clustering methodin the user and project scoring matrix leads the high similarity user cluster into group.Finding the nearest neighbor process is constant in cluster group, which reduces thequery space effectively and improves recommendation real-time.The above method is using the user and project scoring matrix, without combingthe project attributes a new method measure the preference for the project properties isproposed in this paper. The clustering process reduce the query space and mean whileimprove the recommendation accuracy rate. Second, to deal whit the problem of poor recommendation accuracy rate, this paperintroduce a fuzzy improved K-means algorithm cluster in commodity attribute matrix ofproject, and then fusion the similarity the project attach the cluster group fuzzyclustering method and user rating matrix similarity, experiments show that the algorithmaccuracy is superior to the traditional mixed.Third, to solve problem of the user rating matrix is sparse, this paper proposesfuzzy cluster algorithm fuzzy K-means to cluster on user attributes matrix and projectattributes matrix respectively, and taking the average weighted value of two clusteringclusters result as a sparse matrix fill value. The experiments show that this method caneffectively solve the problem of sparse data, it brought a lower accuracy and hasadvantages to other common sparse matrix filled method.Considering the sensitive issue of the initial cluster centers for fuzzy K-means, thispaper proposes a method to calculate the initial cluster centers, the improved algorithmis applied to the above process, the experiment results show that the improved fuzzyclustering still has room for improvement.