Agent-based automated persuasion with adaptive concessions tuned by emotions

Human-to-agent automated negotiation has many potentials in a variety of applications. How to design an agent with equivalent persuasion capabilities with its human rivals is the key to the success of such systems but the research on this problem is still at its early stage. With the aim of improving agents' persuasion ability, this paper proposes to construct emotional agents and emotion-dependent persuasion actions in automated negotiation with multiple issues. First, a multi-issue evaluation function adjusted by the rival's reputation is constructed to determine whether emotional persuasion is needed. Then, by applying the Weber-Fechner Law, this paper proposes a method to measure an agent's emotion generated by evaluating the rival's proposal. Persuasion is categorized into four types and an emotion-based method is proposed for an agent to select a persuasion type. The selected persuasion type is further related to updating concessions, so that an agent can make concessions adaptive to both the rival's proposal and the focal agent's emotional state. Moreover, a series of numerical experiments on bilateral negotiation between agents are conducted to illustrate the proposed model and validate its effectiveness in improving negotiation efficiency. Theoretical and practical implications as well as limitations are discussed in the end.