Personality-guided Cloud Pricing via Reinforcement Learning
Peijin Cong,Junlong Zhou,Mingsong Chen,Tongquan Wei
DOI: https://doi.org/10.1109/tcc.2020.2992461
IF: 5.697
2020-01-01
IEEE Transactions on Cloud Computing
Abstract:As an efficient commercial computing paradigm, cloud computing provides various computing and storage resources to users in a pay-as-you-go manner. However, existing cloud pricing models and mechanisms are deterministic to some degree, thus, may not work well in a real-world environment where user perceived values with respect to cloud services are dynamically changing and highly personalized. In this article, we develop a reinforcement learning (RL)-based dynamic cloud pricing scheme to optimize both cloud provider's profit and costs of heterogeneous users with distinct personalities. Specifically, we first propose a novel personality-guided user perceived value prediction scheme to proactively capture the dynamics of the users' perceived values with respect to cloud services. The prediction scheme models the relationship among user personality, service price, quality of service (QoS), user satisfaction and perceived value in the cloud service market. Second, on the basis of the prediction model, a RL-based cloud pricing mechanism is developed to learn sequential service pricing decision-making for profit and costs optimization. Particularly, the profit and costs optimization problem is modeled as a discrete-time Markov decision process (MDP) that is solved by using Q-learning. Finally, extensive simulation experiments have been conducted to verify our user perceived value prediction scheme and RL-based cloud service pricing mechanism. Simulation results show that our perceived value prediction scheme can achieve up to 87.50 percent prediction accuracy, and our RL-based pricing mechanism can obtain up to 19.39 percent more profit than the state-of-the-art scheme.
computer science, information systems, theory & methods