(An) elicitation-based multi-user preference learning scheme for an IoT-enriched smart space

Abstract

There have been several efforts on preference learning in a smart space. However, most of the works focused on learning single-user preferences while spaces are shared by multiple users in the real world. It is much difficult to provide satisfactory services in a multi-user case than a single-user case since a system need to care about conflicting preferences of them. Only a few works have been proposed a multi-user preference learning scheme in smart spaces. Their systems learn user preferences by observing device-user interaction. However, social psychologists addressed that some users do not reveal their preferences in a multi-user situation due to their personality or social roles. We define a such situation as a potential preference conflict. In this paper, we propose a elicitation-based multi-user preference learning scheme in a smart space. We develop a distributed version of multi-agent reinforcement learning for capturing task-oriented user preferences in a smart space. The system detects any potential preference conflict for improving multi-user conflict situation using asserted harmonization point, which is a preference value where the system believes that targets users would take for a give situation by considering their individual personality, task, and corresponding preference. To yield a harmonized multi-user preference in a smart space, the system explore possible preference candidates considering context sensitivity of user. Experiment results show that multi-user finds the harmonized preference well by using the proposed elicitation-based multi-user preference learning scheme in a short amount of time.