Order Statistics Bayesian-Mining Agent Modelling for Automated Negotiation

The availability of qualitative knowledge has been recently used to simulate human negotiations accurately. During real-life negotiation sessions, people accumulate their knowledge to opt for most adequate bids by which both negotiating parties reach a win-win agreement. Unfortunately, existing research mainly concentrates on few negotiation bids. This paper proposes order statistics Bayesian-mining agent approach to automate bilateral multi-issue multi-session win-win negotiation problems. The proposed agent applies a reallife social bid ranking based on historical bids of all previous negotiation sessions to dynamically update all issues’ weights and preferences. Moreover, it uses our proposed deterministic Trade-Off counter offer method, rather than the existing haphazard estimation method, to estimate precisely the next bid. Experiments are conducted on 3-issue, 5-issue, 6-issue and 10-issue having 27, 3169, 3122 and 13219200 bids respectively. The selected evaluation analysis methods are mainly Pareto optimality, utility, cost and step-wise measurements. Compared with existing agent sorts, such as ABMP, Trade-Off, Bayesian and Mining agents, the proposed agent approach is proved that it is more efficient, effective, scalable and sensitive (adaptable to the opponent steps). Also, it works better to maximize its utilities and to minimize the negotiation costs (the number of rounds).