Designing negotiation agents for automated negotiating agents competition (ANAC)

Abstract

Negotiation is an important process for conflict resolution and forming alliances. Given the increasing interests and advances in automated negotiation in recent years, this paper will discuss one of the participants for the upcoming Automated Negotiating Agent Competition (ANAC) 2015, XianFaAgent. XianFaAgent takes an exploitative stance at the start of a negotiation session and estimates the opponent’s concessive behavior to make adjustments on how much it should concede. XianFaAgent also bases its own concessive function based on the degree of similarity of preferences amongst the competing agents, ensuring that it is not short-change of too much utility in cases where optimal social welfare can be achieved. The agent can effectively handle any linear utility spaces and learning function of the agent is also an important part of the agent’s overall strategy with adaptability in mind. The agent implements two forms of learning, a frequency based strategy for the modeling of the opponents’ preferences and another concessive function that keeps the agent updated before the start of every act to determine which bid is accepted. The results shows that XianFaAgent can effectively negotiate and outbid other ANAC finalists of previous years, although further testing is still recommended and highly encouraged.