Clustering-based Collaborative Filtering Recommendation Algorithm And Application Research

As the development of Internet,the way to access information has changed greatly.On the one hand,the information on the Internet is exponentially increasing,the continuous development of search engine technology and constantly decreasing the threshold of information acquisition provide great convenience for people to find information.On the other hand,for many users,the search engine technology could not help them to search for the useful information easily by one button operation from mass information environment.Especially when people do not provide a good description of the information they are looking for,in other words,when search engine technology doesn't work very well,they can difficultly find the information they need.How to select useful information from mass information and filter out other redundant information effectively has become a hot and difficult topic in the Internet Technology research field.In light of this situation,people began to explore and research the theory and technology of recommendation system.Recommendation system is aimed to help users select out valuable information,and filter out a large amount of useless information,so as to realize the directional push of information.When there are a variety of alternatives for users,recommendation system will help users select the products they preferred.Recommendation algorithm is the core module of recommendation system,which is used to complete the main workflow of recommendation system.In which,the collaborative filtering recommendation algorithm based on clustering process is famous for its superior algorithm performance and higher recommendation accuracy,so it has been applied to many E-commerce systems and portal sites.This paper studied the Collaborative Filtering algorithm based on clustering analysis,and proposes the optimized and improved scheme according to the performance of the existing clustering algorithm in recommendation system and its advantages and disadvantages.And add the new clustering process into the recommendation process,and then obtain the higher accuracy recommendation results.Semi-Supervised PSO clustering algorithm balances the advantages and disadvantages of unsupervised PSO and standard PSO clustering algorithm by manually set a parameter marked ? Since the selection of parameter ?has an effect on the accuracy of clustering results,then we proposed an algorithm with adaptive parameters to improve the selection process of parameter ?.This paper combines the PSO algorithm with Semi-Supervised Learning,and studies the selection method of parameters for Adaptive Parameter Optimization Semi-Supervised PSO.Then propose Adaptive Parameter Optimization Semi-Supervised PSO clustering algorithm(APO SSPSO).The basic idea of the algorithm is to optimize the parameter which represents the distribution proportion of labeled samples and unlabeled samples and obtain its best value.Then advance the learning ability for the selection of this parameter,so as to improve the clustering effect of Semi-Supervised learning and the accuracy of the recommendation results.This paper obtains some user groups which have relatively low clustering error by using APO_SSPSO.In the clustering result,the users in the same cluster have high similarity,and the users in different clusters have low similarity.Then the following recommendation equation could select the optimized nearest neighbors.At the same time,we can optimize the proportion of local neighbors and global neighbors in the recommendation equation by using PSO algorithm again to further improve the accuracy and recall rate in recommendation results.We will make a comparison between APO_SSPSO User-CF and traditional User-CF,and then validate the validity for the algorithm we proposed.We evaluate the performance of APO_SSPSO User-CF and traditional User-CF by using the index of accuracy,recall and coverage.