Study On Collaborative Filtering Recommendation Algorithm Based On Clustering

In collaborative filtering recommender systems,products are regarded as features and users are requested to provide ratings to the products they have purchased.By learning from the ratings,such a recommender system can recommend interesting products to users.However,there are usually quite a lot of products involved in E-commerce and it would be very inefficient if every product needs to be considered before making recommendations.The clustering algorithm has also received much attention in the recommender systems in recent years because of the inherent advantages of clustering itself,which can solve the problem of excessive data dimension with a large extent.Based on the characteristics of ItemRank algorithm and the advantages of clustering algorithms,this paper proposes an improved ItemRank algorithm,IRSCC,which can improve the efficiency of ItemRank algorithm effectively.The improved algorithm proposed in this paper,is applies a self-constructing clustering algorithm to reduce the dimensionality related to the number of products,Recommendation is then done with the clusters.Finally,re-transformation is performed and a ranked list of recommended products is offered to each user.With the proposed approach,the processing time for making recommendations is much reduced.Experimental results show that the efficiency of the recommender system can be improved without compromising the recommendation quality.The main research work of the improved algorithm of this paper is mainly in the following aspects:(1)Through the research and analysis of the advantages and disadvantages of ItemRank algorithm,and then use the clustering technology to improve the shortcomings of ItemRank algorithm.Reference to the current better clustering algorithms,excluding the K-means and other clustering algorithms,which has a large burden to the user or perform time-consuming,finally use a self-constructing clustering algorithm,which has used in the area of the text mining and does not need to set the number of clusters in advance,to improve the ItemRank algorithm.(2)The improved algorithm proposed in this paper,IRSCC,has five main steps: first of all,in order to facilitate the reduction of products,we assign a class label for the user.Similar users are clustered into the same cluster,and dissimilar users are clustered into different clusters.Finally,users in the same cluster have the same and unique class tags.Before labeling tags,we eliminate the impact of the user rating scale on the recommended results.Then the SCC algorithm is used to reduce the dimensionality related to the number of products.Before the reduction of the dimensionality related to the number of products,we firstly use a method,proposed by Jiang et al.,to construct the characteristic pattern which conforms to the feature of the dimensionality reduction algorithm,and then reduce the dimensionality,compress the original data set.Secondly,we create a correlation graph with the reduced data set which shows the inter-relationship among the q product groups.This step is the same as the first step of ItemRank algorithm,but it has been adjusted.Then a series of random walks are performed to get the derived preference list of product groups.Finally,we transform the list of product groups to preference list of the products for user,then recommend them to the user.(3)Using the open and accepted experimental data set,the improved algorithm proposed in this paper has been simulated and tested.Then the experimental results are compared with the ItemRank algorithm and other collaborative filtering algorithms,which using dimensionality reduction techniques and graph techniques.The rationality and effectiveness of the improved algorithm are verified by the experiment results.Finally,the experimental results show that RSCC algorithm can effectively improve the efficiency of ItemRank without compromising the recommended quality,and it is also more effective than other recommender algorithms based on clustering techniques.The experimental results were analyzed in detail.The reason of the difference between the actual and the theoretical appreciation value is analyzed by comparing the effect of the implementation of the algorithm.