Partner Selection Model For E-commerce Muti-Agent Negotiation Based On Learning And Trust

As the flourishing development of e-commerce and the maturation of software technology, the intelligent Agent presented a great application prospect in e-commerce due to the inherent characteristics of autonomy, sociability, and interactivity. Agent can inherit humans' preference and interests, perform many key steps of e-commerce rapidly and efficiently, and finally achieve the purpose of atomization, intelligence and personalization. For this reason, Agent-Mediated Electronic Commerce (AMEC), become one of the most important research topic both in e-commerce and artificial intelligence.Agent was typically interaction with other agents by some form of negotiation, and then execute the transaction according to the attribution on the contract, but for the dynamics and openness of the market, Agent may face many negotiation partner simultaneously, which has different preference, profit, believe and transaction objects, so agent have to choose the most appropriate one by with certain strategy before the negotiation thread, it has a great meaning for the improvement of success rate of and efficiency and transaction, increasing of user's transaction gain and the operation efficiency of e-commerce system.This paper proposed a partner selection method based on learning and trust after summarize of the deficiency of existing literature. First, through the analysis of the interactive property of negotation,we introduce the support vector machine to negotiation decision process and achieve the purpose of negotiation history learning, and then we integrated the leaning process into a simulate negotiation process, through which we predict the probable outcome and finally negotiation profit of current negotiation; besides, we also developed a probability-based trust model aiming at the AMEC process of transaction and evaluation, we first designed a method for transaction result evaluation in AMEC, with this method we get the initial ratings which was used as the primitive information for trust evaluation, we developed a mechanism for the determining of trust accuracy based on probability theory, follow with a new model of recommendation reliability based on the thinking of hypothesis test, it is a combination of trust accuracy, recommendation consistency, recommendation efficiency, then we make a good aggregation of trust information with these component; at last, we combine negotiation profit and trust appropriately, which were used for the finally prediction of transaction profit; we also conducted a simulated experiment, by which we illustrate the process of partner selection with our methods, and analysis the performance of many parameters and the whole model.