Research On Personalized Recommendation Method Based On User Clustering

Recently,the rapid development of Internet technology makes Web become an important way for people to obtain information,however,"information overload" makes people lost in the increasingly rich ocean of information and diverse types of information,the recommendation system can help users to effectively solve the network information lost.The application of clustering technology in recommendation can alleviate the problem of data sparsity,scalability and cold start,but the introduction of clustering technology not all can improve the results.This is due to the clustering method itself has some defects,such as the clustering quality is not high,the clustering results are unstable,The application of the clustering results to the recommendation will inevitably lead to the decline of recommendation quality.How to apply the clustering technology to the recommendation system in order to improve the accuracy of the recommendation results is the focus of the research based on clustering.The idea of this study is to establish a user clustering model,which can be used to solve the problem of data sparsity in collaborative filtering and distribute non-uniform initial resource values in the mass diffusion algorithm.Based on this idea,two recommendation algorithms are proposed,merging clustering user ratings of collaborative filtering recommendation algorithm(UCCF)and non-uniform resource distribution mass diffusion recommendation algorithm based on user clustering(UCMD).The proposed algorithm mainly uses the clustering technology,the user clustering model is introduced to recommendations.The main research contents of this thesis include the following three aspects.(1)In order to solve the problem of data sparsity in traditional collaborative filtering recommendation,a novel merging clustering user ratings of collaborative filtering recommendation algorithm is proposed.First of all,clustering technology is used to cluster the users,and the other users in the cluster where the target user is located are used as the nearest neighbors,merging the neighbor ratings of the cluster can be used to generate new ratings,which are filled into the original rating record.The similarity between users is calculated by the new rating data,In this way,the nearest neighbors are more accurate than the traditional method.The prediction rating is calculated by the accurate neighbor users is more accurate.(2)The method of setting the initial resource value of the object to 0/1 in the recommendation algorithm of two graph network structure,non-uniform resource distribution mass diffusion recommendation algorithm based on user clustering is proposed.First of all,clustering technology is used to cluster the users,according to the clustering model,the objects selected by the target user and the objects selected by the users with the target user in the same cluster are set with different initial resources,Then the classical mass diffusion algorithm is used for the following recommendation.(3)The proposed UCCF algorithm and UCMD algorithm are tested on real data sets on Movielens sites.The experimental results of UCCF algorithm show the use of collaborative filtering algorithm in the data after the user clustering model is used to fill the ratings and the use of a collaborative filtering algorithm in original rating,the mean absolute error(MAE)of the former is much lower.The experimental results of UCMD algorithm show the algorithm is superior to the standard mass diffusion in precision,rank score and inter-user-diversity,novelty is the same as the standard mass diffusion algorithm,The precision of recommendation is improved as well as inter-user-diversity,and novelty is also maintained at a high level.