Stepwise Clustering And Similarity Optimizing Based Collaborative Filtering Algorithm

Nowadays,collaborative filtering technology is a kind of personalized recommendation technology that is widely and successfully used in the personalized recommendation system.But with the expanding scale of the recommendation system and increasing amount of internal data in practical application,this technology has exposed a lot of weaknesses such as data sparsity,poor scalabality,low precision and so on.To solve the weaknesses above,a collaborative filtering algorithm which is based on stepwise clustering and similarity optimizing is proposed in this paper.First,to deal with the data sparsity and poor scalability which are existed in collaborative filtering technology,this thesis proposed a stepwise clustering strategy on the basis of collaborative filtering algorithm based on clustering.An improved K-means algorithm is referred to avoid initial cluster centers are selected too close which make the algorithm involve into a local optimal solution during the stepwise clustering processing for projects and users.The stepwise clustering improves the scalability of the algorithm,and relieves effect of the sparse matrix on the clustering results.Second,to avoid low Precision is caused by the directly calculation of similarity in collaborative filtering algorithm,A mathematical model of similarity optimization among users is presented that is based on the score prediction formula in this paper.And the score data are used to train and optimize the similarity and that’re getten from stepwise clustering.This method is proposed without losing the scalability while improving the accuracy of the similarity computation,finally improving recommendation accuracy of this algorithm.At last,in order to verify recommendation performance of the proposed algorithm in this paper,we conducted experiments and analyzed the experimental results.The results show that compared with the traditional Collaborative filtering algorithm,Item-based and User-based collaborative filtering algorithm,the improved algorithm in this paper improves the recommendation accuracy of collaborative filtering algorithm,and has a better recommendation quality.