A modular approach to landmark detection based on a Bayesian network and categorized context logs

Mobile context logs contain meaningful and private information about their owners that can be used to support users' human memory. However, it is difficult to efficiently retrieve the information because of the enormous amount of mobile context logs and the limitations of mobile devices in terms of power, memory capacity, and speed. To efficiently retrieve information, detection of important events or landmarks is required. In this paper, we propose a modular approach of a Bayesian network for landmark detection using categorized context logs. The proposed model consists of several modules of Bayesian networks used to reduce the time of inference and the size of memory used, and each module is learned using categorized context logs according to the days of the week in order to decrease learning time and increase accuracy. Our experiments on Nokia log data and our life-log data show that the modular approach is superior to a monolithic Bayesian network and confirm that using categorized context logs for learning enhances the inference performance.