User behavior modeling for AR personalized recommendations in spatial transitions

Abstract

There have been studies on personalized augmented reality (AR) systems taking users' contexts and histories into account. However, there is insufficient research on incorporating real-time user behavior and interactions into the personalized recommendations in spatial transitions, which can be used for new users without user history data. The spatial transitions, distances between two Point of Interests (POIs) of an AR tour trajectory through which users should pass to visit AR contents, need to be filled using the personalized AR contents to reduce the discontinuity experience. This paper aims to propose a user behavior model to recommend personalized contents in the AR tour trajectory to create a personalized experience. First, we model three interactions, including staring, skipping, and liking, using mathematical methods; and two spatial behaviors, including standing and moving, using a rule engine. Second, a content filtering-based recommendation method is presented to apply the proposed user model to recommend personalized AR contents in the spatial transitions. The experiment results showed that the proposed real-time user behavior model brought notable improvements in creating a personal experience of the AR tour system. The proposed method outperformed the conventional method in terms of recommendation performance metrics, so that it achieved an average of 15% more precision and recall scores and an average of 32% more personalization scores than those of the conventional method. Furthermore, the level of personalization perceived by the proposed user model demonstrated significant positive relationships with the level of decision effectiveness, system trustworthiness, and perceived recommendation transparency. The findings showed the effectiveness of the proposed framework in solving the cold-start recommendation problem.