Abstract: Web information is scaling more than exponentially with time. How to acquire information efficiently by personal search engine is staring us in our faces. Personal preference can not be easily described but can be learned quickly from the examples. Although PCC (pairwise classification clustering) is a powerful tool for learning the examples, but transitive dependences dwarf it. In this paper, we introduce clustering with SVM and define semantic cosine similarity based ontology to solve this problem. Experiments proof that it is efficient and powerful.