With the exponential growth of World Wide Web, it becomes more and more time and energy consuming for users to find what they're interested in. It leads to a clear demand for personalized services, which can provide different users with different services User modeling is the key technology in implementing personalized services. Conventional user modeling methods with both positive and negative examples will either interfere users' normal browsing or bring in noises. A user modeling method without negative examples is presented. A hybrid of genetic algorithms and kNN classifier are utilized to search the words describing users' interests. The method is applied in personalized information filtering. The experiments show that the filtering ratio and precision can be 73.91% and 94.44% respectively, which demonstrates that our user modeling method is feasible and efficient.