隨著電子商務的蓬勃發展,學習型代理人協商變得越來越重要。買賣雙方透過協商可以找出合理的價格,然而要達到買賣雙方共同利益最大化還有一段距離。本研究以協商決策函式(Negotiation Decision Function)為協商基礎,提出一雙邊協商、多議題、單方學習的學習型代理人之協商模式,此學習模式是透過類神經網路的學習機制,來學習對手的讓步戰術及各議題的權重值大小。本文中探討原始法模式、正規劃模式、比值法模式等三種學習模式,並針對10種不同協商次數來作實驗,以分析比較出最佳的學習模式。在本研究的結果中,三種學習模式以比值法模式的學習效果最佳,不但精準猜測出對手的偏好,且協商次數最少。另外,針對航空貨運承攬業為例建構一多目標貨物排程模式,以規劃出一最佳的貨物運送路線,以達到成本最小化及運送時間即時化的目的。最後,將學習型代理人的協商模式運用於承攬業的貨物排程上,並在協商過程使用權衡機制來使協商結果能夠接近柏拉圖最佳化。
When the Internet is getting popular, adaptive agent negotiation becomes more important. Buyer and seller deal a rational price by negotiation, but it is not an optimal solution for both sides. The study uses the Negotiation Decision Function to develop a bilateral, multi-issue, and single adaptive negotiation model. The model is to adapt to the concession tactics and the weight of every issue offered by the opponent. The study investigates three adaptive models: original data method, normalized data method and data ratios method with an experiment of ten types of different negotiation times in order to determine the optimal adaptive model. In the study, the data ratios method has the best adaptive performance among these three adaptive models, and it can exactly guess opponent’s preferences for less negotiation times. In addition, this study takes air freight forwarder scheduling as an example to develop a multi-objective model. It can solve an optimal route of transportation for minimizing both cost and transportation time. Finally, the study applies the adaptive agent negotiation model to the case of air freight forwarders scheduling, adding the negotiation trade-offs mechanism to the process of negotiation in order to achieve a Pareto optimal solution.