(An) efficient multi-user preference learning scheme in smart spaces using task precedence weights and incremental preference differentiation

Abstract

We address the problem of preference learning in multi-user smart spaces in the viewpoint of distributed multi-agent system and algorithm. As multiple users may have different and conflicting preferences, the preference learning system should be able to understand the resolution methods so that it can provide appropriate services to groups. Moreover, due to the large number of user groups, an efficient learning scheme is needed to deal with the group preference learning. We suggest a method that utilizes task-dependent weights that resolve the conflicts and are learned over time according to the users' behaviors. Also, we propose incremental preference differentiation scheme that is useful to manage large leaning space and speed up the learning of similar preferences. We have successfully evaluated the proposed system and algorithm by a simulation experiment in a testbed equipped with eight smart objects.