Applying Reinforcement Learning in Context Limited Environments

Abstract

Reinforcement Learning has seen some interesting development over the last years, which made it very attractive to use on recommendation scenarios. In this work, we have extended the previously developed pervasive system, which is aware of the conversational context to suggest documents potentially useful to the users, with the ability to use users’ click data as a way to perform better suggestions over time, through a Reinforcement Learning approach. Furthermore, to assure the real significance of these types of approaches in conversational environments, we also conducted a case study regarding the accuracy of feedback on context limited conversational systems.