Abstract: For the growing of the status of Web data mining, this paper proposes a new approach of Web mining based on support vector machines and clustering, and taps new methods, according to the theory of correct divided areas that support vector will only appear in the interval of two types of sample collection, through the introduction of concepts of such as clustering center of mass, clustering radius and clustering centroid distance, thus resolves better the problem of fast and accurately remove non-support vector to ensure the generalization of algorithm. The experimental results show that this improved algorithm not only can fast and precisely delete the training samples but also has a better solution to the issue of generalization.

Key words: Web mining, support vector machine, clustering