The emergence of open mobile platform accelerates the development of service composition in mobile application. However, most current mobile applications fulfill service through static and clustering-based composition. It is prone to "software function overloading" and overlong service accessing path, hindering the efficient use for the users. In this paper, an exploratory service composition method for mobile application. With this method, candidate composable services are provided to the user when the context changes and the selected service(s) is(are) composed into the application just in time. To acquire the candidate services, association rules between the contexts and services are mined based on the history of service selection by different users in different contexts, and used as the matching basis. Further, the traditional FP-tree algorithm is extended to enable the mining of two-dimensional items in mobile application. Experiments show that the extended algorithm has higher precision and recall rate than the traditional method in matching for candidate services.