Abstract
In order to increase the chance of successfully achieving an agreement in negotiation dialogues, agents should be certain about their moves selection. Negotiation in ambiguous conditions such as incomplete information settings, leads to uncertainty about the played moves and makes the task of making decisions more complex. In this paper, we study and analyze agents' uncertainty in argumentation-based agent negotiation (ABAN). We present two methods: Method I, and Method II. Method I extends a previously defined approach for measuring agents' uncertainty based on Shannon entropy. Method II introduces a new technique to assess agents' uncertainty using hypothesis testing. The objective is to build a comprehensive framework for tackling agents' uncertainty in ABAN. Furthermore, we propose a Markov decision process-based framework to incorporate the different situations of agents' uncertainty based on the available arguments or offers. Besides, we discuss the uncertainty issues in two special cases based on the different classes that arguments can belong to. In addition to the theoretical analysis of arguments uncertainty, we discuss the implementation of the proposed approach by applying it to a concrete case study (Buyer/Seller) scenario. The obtained empirical results confirm the effectiveness of using our uncertainty-aware techniques and show that our negotiating agents outperform others that use pure argumentation with no uncertainty consideration .