Phoenix: A threshold function based negotiation strategy using gaussian process regression and distance-based pareto frontier approximation

Abstract

Automated negotiation is of great interest in artificial intelligence. An effective automated negotiation strategy can facilitate human in reaching better negotiation outcomes benefiting from the adoption of advanced computational methods. This paper deals with multi-lateral multi-issue negotiation where opponents’ preferences and strategies are unknown. A novel negotiation strategy called Phoenix is proposed following the negotiation setting adopted in The Sixth International Automated Negotiating Agents Competition (ANAC 2015) [13]. In attempt to maximize individual utility and social welfare, we propose two highlighted methods – Gaussian Process Regression (GPR) and Distance-based Pareto Frontier Approximation (DPFA). Integrating the idea of these methods into a single function called threshold function, we show that Phoenix is a fully adaptive, cooperative and rationally designed strategy.