A mobile network has higher demands for the performance of personalized mobile network services,but existing researches have been unable to modify the contextual mobile user preferences adaptively and providereal-time, accurate personalized mobile network services for mobile users. This paper proposes a contextcomputing-based approach to mobile user preferences adaptive learning, which can ensure the accuracy and theresponse time. First, through analyzing the logs of contextual mobile user behaviors, the method judges whethermobile user behaviors are affected by context or not, and detects whether the contextual mobile user behaviorschange. According to these judgments, the contextual mobile user preferences are modified. Secondly, the context isintroduced into the least squares support vector machine (LSSVM), which is employed to learn the changedcontextual mobile user preferences. Further, a learning method of contextual mobile user preferences is proposedwhich is based on context of the least squares support vector machine (C-LSSVM). Finally, the experimental resultsshow that the proposed method is superior to other learning methods when considering both accuracy and responsetime. The proposed method in this paper can be applied in the system of personalized mobile network services.