A similarity-based learning approach for adaptive negotiations

Negotiation is a crucial step in the process of multi-agent decision making. Theories and techniques for automated negotiations have substantial practical values for agent-mediated electronic commerce. As a negotiation context tends to change over time, negotiation agents must be a to learn the changing contextual information (e.g., current preferences of their opponents) in order to make sensible deal acceptance decisions and to speed up the negotiation processes. Existing adaptive negotiation methods are still primitive in terms of what a negotiation agent can learn (e.g., price only) and how responsive an agent is towards the changing negotiation issues. This paper proposes a novel similarity-based learning method for adaptive negotiation agents. These agents are sensitive to multiple issues in a changing negotiation context. By observing their opponents' moves, these adaptive negotiation agents can make more sensible counter offers to speed up the negotiation processes. According to our preliminary experiment, the proposed similarity-based learning negotiation agents outperform their non-adaptive counterparts. In addition, their performance is comparable to that of the more sophisticated genetic algorithms based adaptive negotiation agents.