Personalized Recommendation Of Particle Swarm Optimization Based On K-means Clustering

With the rapid development of e-commerce, the amount ofinformation in the network increases greatly, how to identify and discoverthe information to meet user’s needs, which is the key problem ofresearch in data mining field. As an important branch of data mining, Webmining emerged, its aimed to find the user’s behavior patterns and get theuser’s real needs by analyzing Web log files, which providingpersonalized recommendation service.This paper analyzes the definition, process, the characteristics ofWeb data mining, and compares the related techniques, put forward theimprovement measures for the inherent defects of personalizedrecommendation algorithm, effective solve the efficiency ofrecommendation in real-time recommendation process. Experimentsshow that the improved algorithm is reasonable.This paper focus on the off-line clustering, in order to enhancereal-time of recommendation algorithm, using the off-line clustering ofWeb log, clustering algorithm using K-means, the smaller the target pitchbetter clustering, and can handle large data sets, but the algorithm’sdefects are also obvious, on the one hand the mechanism of algorithm todetermine the high cost of time, on the other hand, the selection of theinitial cluster centers have enormous impact on for the clustering results.In order to solve the above two problems using off-line clustering in thispaper, that is, in advance of the processing for the log files, and theintroduction of chaos particle swarm algorithm, particle swarmoptimization algorithm because of its simplicity, few parameters to set,can accelerate convergence of the algorithm, but it is very easy to fall intolocal optimum, therefore the global search ability is poor, the uniqueergodicity of chaotic sequence can effectively help the particle escapefrom the shackles of local optima and improve the global search ability.Experimental results show that the clustering effect is better.Despite the off-line clustering have obvious advantages, the clustereffect is better, but the recommended algorithm requires real-timeprocessing, how to coordinate the relationship between online and offline is an important research direction in the future.