Emotional agents enabled bilateral negotiation: Persuasion strategies generated by agents' affect infusion and preference

Abstract

Agents with emotional intelligence can enhance negotiation outcomes by improving interaction and better understanding of human opponents in human–machine negotiation. This paper proposes a novel method to improve agents' emotional intelligence level by modeling emotion and preferences in negotiation. Emotion is modeled to reflect an agent’s enduring emotional state and its dynamically instantaneous response to the opponent’s proposals. Emotion-driven rules are subsequently constructed to generate persuasion strategies. Then, a two-step issue updating model which integrates persuasion strategies, attribute preference, and altruistic preference is established to generate optimal issue values using the simulated annealing algorithm. After that, the proposed model is validated by agent-agent experiments and human-agent experiments. The experimental results show that (1) the introduction of emotion and persuasion strategies enhances negotiation success, with joint utility rising by 11%, utility difference decreasing by 28%, and negotiation speed rising by 24%.; (2) the inclusion of preference reduce the utility difference and improve the negotiation fairness, with attribute preference and altruistic preference reducing utility difference by 12% and 18%, respectively; (3) the proposed model outperforms competing models in negotiation speed and joint utility, and ensures larger joint utility and minimizing differences even when opponent's preferences are unknown. The results imply that the proposed model is conducive to achieving a win–win human-agent negotiation outcome.