During the process of service retrieval, it?s often dif-ficult for users to give exact retrieval requirements be-cause users are not familiar with the complex description mechanism of services. This limits the function of ontol-ogy service models and leads to lower completion, preci-sion, efficiency and easiness of service retrieval. It is ur-gent to have an efficient method to help the users. The paper introduces a selfadaptive learning algorithm based on association mining theory in data mining field to learn from the retrieval history and assist users in giving high quality retrieval requirements. The experiment results show the effectivity of the proposed algorithm.