The countless amount of knowledge is produced everyday. As the effort of a user to find the relevant information with his or her interests gets bigger, so does the importance of personalized and intelligent information delivery. In knowledge portal or knowledge management system(KMS), the basic concept of recommending knowledge is derived from information filtering. But it is hard to apply information filtering to knowledge portal directly, because the basic information filtering algorithm assumes that filter relevant information with a single domain of information. But a user usually has more than an interest in a knowledge portal; the knowledge recommendation system should filter these unrelated interests simultaneously. Instead of containing all interesting information for a user in a single profile, dividing this user profile into several sub-user profiles according to the domain of information will increase the effectiveness of information filtering. For this, I suggested a clustering algorithm which binds interrelated documents and separates unrelated documents to compose a multi-user profile. Each of multi-user profiles can grasp more relevant information and adjust the number of sub-user profile according to the level of interests for a user without fixing the number before clustering. Furthermore this multi-user profiles enable the selective learning of user interests according to the dynamics of a sub-user profile. In other words, a system can adapt to the interest change of a user better than the general information filtering approach. The performance of suggested algorithm is verified by conducting an experiment and a survey.