Design And Implementation Of Coalitional Negotiation Platform And Agent Negotiation Algorithm

Artificial intelligence is concerned with intelligent behavior in artifacts,such as defeating humans in strategic encounters or outperforming humans in cognitive tasks.Coalitional games can be used to model problems arising in a wide variety of domains such as economics,politics,and many other social fields,which are a type of mixedmotive(both competitive and cooperative)game where a group of players strategically negotiate to decide what coalitions to form and how to divide payoff among coalition members.Although there exists a large body of work on coalitional game theory,the process of the coalition formation has received less attention,especially when agents interact with humans rather than with other autonomous agents.A main challenge for AI negotiating with people in such games is that the outcomes not only rely on negotiation strategies,but also on intuition,cultural norms,signals and emotions.This work therefore designs and implements the online negotiation environment for coalitional game,which aims to provide a user-friendly negotiation platform for training players' negotiation skills and developing agent-based automatic negotiation algorithm.This work also presents a novel approach combining game-theoretic theoretical solution concepts,machine learning models with mechanisms for generating and acting on signals to produce a new learning algorithm that deals with people and other agents in coalitional negotiation games.Several agents are also created following competing philosophies.We evaluate the performance of those agents in the online coalitional negotiation platform,and the results suggest several important properties for agent design in such contexts.