This paper describes how collaborative filtering can be used for mobile devices. When the user is connected to a central repository, the algorithm selects a subset of profiles to store on the device. When the user is not connected to the repository, the predictions can be incrementally updated to reflect new or updated ratings. Experiments on a movie data set show that the method can dramatically reduce the data needed while still performing nearly as good as a centralized approach.
Download Full PDF Version (Non-Commercial Use)