Research On Recommendation Algorithm Based On Clustering

Recommendation system and clustering are two widely used data mining techniques.Recommendation system can help users obtain useful online resources efficiently and effectively.Clustering technology can reduce the high dimension and sparsity of data,cluster similar objects together as much as possible,and separate different objects.In recent years,it has become a trend to group users or items with similar characteristics before making recommendations to improve the quality of recommendations.In this paper,from the point of view of optimizing the disadvantages of common clustering and eliminating the key parameters of clustering,we put forward the idea of combining clustering with recommendation algorithm,that is,collaborative filtering algorithm based on clustering and simulated annealing and optimal user recommendation algorithm based on hierarchical clustering tree.1)Collaborative filtering algorithm based on clustering and simulated annealing:First of all,we according to the user to watch the movie movie category,the user behavior into history and film categories related user type vector,then user type vector as clustering objects,classify the users with K average clustering algorithm,in which using simulated annealing algorithm is optimized,finally respectively using collaborative filtering algorithm for the unit with the clustering of the user is recommended,The accuracy and stability of the recommended results were compared by using simulated annealing algorithm.The user type vector and the user history score are used to calculate the similarity between users,and the appropriate proportion of the two similarity degrees is determined through experiments as the final similarity between users.2)The best user recommendation algorithm based on hierarchical clustering tree:Firstly,hierarchical clustering of users is carried out by Ward's hierarchical clustering method,and then the objective function of a single clustering node is defined by the clustering order difference and the variance inside the clustering node.Finally,the Framework For Optimal Selection Of Clusters(FOSC)was used to extract Clusters from hierarchical clustering tree,and different recommendation algorithms were used For verification on different data sets.Based on the above research,can be seen from the performance evaluation,this paper presents the simulated annealing algorithm can greatly eliminate K average clustering algorithm,the initial clustering center of the recommended results brought by the random instability problem,hierarchical clustering algorithm can avoid the clustering "clustering number" key parameters,and is suitable for large data sets,improved the suitability.Both of them belong to the pre-processing stage of clustering as a recommendation system and are representative clustering algorithms.