Integrated data mining methodology for context-aware recommendation based on dynamic user behavior

Abstract

The evolution of mobile and context-aware computing environment enables mobile service providers to obtain various situation information which has not been available before. Using this situation information, it is possible to provide the more tailored applications that usefully adapt to the environment of the users. Today, mobile service providers have a more direct recommendation channel, namely the short messaging service. Therefore, mobile service providers should consider both the timing and context of recommendation messages (push messages) that are sent to users. Mobile service providers can learn context-specific user preferences by analyzing mobile web use logs and user responses to push messages. In this paper, we describe the user preference model for learning individual preference and present a context-sensitive recommendation system that can be used to select the optimal context in which to send recommendation messages. In order to decide the optimal context for each user, user-service-context preference and service-context preference are extracted from service usage log and they are combined based on the clarity of preference. We compared user responses to push messages delivered in and out of suitable contexts as determined by recommendation system. The precision of push messages delivered within a suitable context was higher than that of messages delivered outside of one.