Learning Algorithms for Software Agents in Uncertain and Untrusted Market Environments

Thomas T. Tran, Robin Cohen

Abstract: The problem of how to develop algorithms that guide the behaviour of personal, intelligent software agents participating in electronic marketplaces is a subject of increasing interest from both the academic and industrial research communities. Since a multi-agent electronic market environment is, by its very nature, open, dynamic, uncertain, and untrusted, it is very important that participant agents are equipped with effective and feasible learning algorithms in order to achieve their goals. In this paper, we propose algorithms for buying and selling agents in electronic marketplaces, based on reputation modelling and reinforcement learning (RL).