Abstract
A context-aware service in a smart environment aims to supply services according to user situational information, which changes dynamically. Most existing context-aware systems provide context-aware services based on supervised algorithms. Reinforcement algorithms are another type of machine-learning algorithm that have been shown to be useful in dynamic environments through trialand-error interactions. They also have the ability to build excellent self-adaptive systems. In this study, we aim to incorporate reinforcement algorithms (Q-learning) into a context-aware system to provide relevant services based on a user???s dynamic context. To accelerate the convergence of reinforcement learning (RL) algorithms and provide the correct services in real situations, we propose a combination of the Q-learning and case-based reasoning (CBR) algorithms. We then analyze how the incorporation of CBR enables Q-learning to become more efficient and adapt to changing environments by continuously producing suitable services. Simulation results demonstrate the effectiveness of the proposed approach compared to the traditional CBR approach.