Web User Clustering Research Based On Improved Shuffle Frog Leaping Algorithm

With the rapid development of Internet, users not only use web to exchange resource, but also use web to share information. At the same time, a mass of data is produced by large-scale of users’ accessing behavior. On one hand, massive amount of web information provides to users a source of information access, On the other hand, how to efficiently obtain valuable information from a mass of web information has been a serious challenge for web users. In practice, the ultimate objective of web user clustering is mining valuable information from massive amount of users accessing log information, thus provide superior service for users. At present, web usage mining technologies have been successfully used in business intelligence, improvement of web site topology, personalized service and so on. As a research plot of web usage mining, the aim of web user clustering is to discover similar behavior users,which can be used to give some help for personalized service, improvement of web site topology and soMost of web user clustering researches are using traditional clustering method to solve web clustering problem. However, when faced with large-scale web user clustering problem, traditional methods can only produce a local optima and become powerless, such as K-medoids. For this case, there are a lot of swarm intelligence optimization techniques applied to solve this issue, such as PSO and ACO. As a algorithm of intelligent bionic optimization, shuffled frog-leaping algorithm(SFLA) has been successfully applied to function optimization, production scheduling and combination optimization by many scholars, for its concept is simple, it is easy to implement,and its parameters are less. However, research on web user clustering based on SFLA is still at the beginning. Therefore, it has great significance to improve SFLA according to the problems characteristics.The main work of this paper includes the following aspects:(1) the calculation of users similarity. The measurement of users clustering feature similarity can directly affects users similarity when calculating the similarity of users. In the research of user similarity, the paper mainly considers user clustering feature similarity from user browsing interest and the order that page is viewed, then calculate user similarity based on this.(2) The improvement of SFLA. SFLA has been successfully applied to function optimization, and production scheduling and combination optimization, but its speed of convergence rate is slow, and it is easy to fall into premature shortcoming. The design an adaptive step moving updating rules for the shortcoming that convergence rate is slow. Adaptive step moving updating rules is conducive to dynamically adjust the bad frog learning from the better frog in local search, thus accelerate algorithm speed of convergence rate. The paper introduces population elimination mechanism to prevent prematurity for algorithm easy to fall into premature phenomenon.(3) the fusion research of improved SFLA and K-medoids. The paper proposes a hybrid algorithm based on the fusion of improved SFAL and K-medoids by analyzing their characteristics, and uses the hybrid algorithm to solve web user clustering.(4)experimental analysis. The paper uses K-medoids and the new proposed hybrid algorithm to cluster web users respectively and analyze the result of clustering. The result of experiment shows that the hybrid algorithm based on improved SFLA and K-medoids proposed in this paper has a better clustering quality than K-medoids.The research of the paper can effectively implements web user clustering based on web log. It also plays an active role in the development of web user clustering and SFLA.