電子化協商之期望效用、戰略及結果預測之研究

Abstract

近年來，電子商務的應用已遍及各種商業領域，從簡單的商品零售到複雜的動態夥伴關係建立，均可見其足跡。為了讓屬性迴異的參與者都能獲得滿足，電子商務也導入了不同的交易模式，其中最重要的包括標價出售(posted-price selling)、拍賣(auction)與協商(negotiation)等三種。在上述模式中，協商顯然是最複雜的一種，也是某些交易中不可或缺的步驟。雖然協商在商業行為中佔有重要地位，但很遺憾地，目前並沒有證據顯示協商模式已被成功導入電子商務中。因此，為進一步拓展電子化交易，如何協助人們在電子商務架構下進行協商仍有待進一步研究。 對於電子化協商(e-negotiation)的推廣，最重要的莫過於發展適當的軟體以協助使用者進行事前準備以及出價過程的指引。針對此目標，發展多年的電腦輔助協商支援系統(computer-aided negotiation support systems)可被視為最佳輔助工具。尤其，目前的電腦具有高速運算能力與大量儲存空間，使得協商支援系統有時甚至可提供比人類智庫更快、更有用的建議。然而，只依賴電腦的高速運算並無法大幅提昇支援品質，事實上，發展有效的支援方法才是將協商支援系統與電子化協商成功結合的關鍵因素。 為提供有效的協商支援，本論文針對以下三項重要主題進行研究: (一) 如何協助參與者在協商前訂定其合理的期望成交效用(aspiration utilities)? (二)如何協助參與者在協商前與進行中預測協商的結果? (三) 如何協助參與者挑選適當的戰略(tactics)與策略(strategies)以達成預設目標? 首先，針對第一項主題，本論文以加總計分模型(the additive scoring model)為基礎提出了一種新的解析方法，讓協商者可根據平均成交機率訂定客觀且精確的目標。當考慮雙人多議題協商的期望效用時，本研究首先提出符合問題本質的限制式，接著以窮舉式模擬實驗(simulation-based analyses)來進行求解。實驗結果顯示: 在未事先假設雙方喜好的前提下，決定協商者的期望效用範圍是可行的。而且，當把一方的喜好視為已知時，期望效用值更可被限制在更小範圍內。 其次，為了預測雙人多議題協商之結果，我們首先證實了協商空間(the settlement space)與成交可能性、成交總效用間有正向之相關性。接著，本研究又分析出雙方的議題權重(weight)差值總合與協商空間亦為正相關。根據上述結論，我們便可利用事先估計的議題權重差值總合來預測協商的結果。 對於第三項主題，本研究以Faratin等人提出的協商模型為基礎，詳細驗證並比較各種戰略的效能，以提供協商者運用時之參考依據。首先，在單議題協商環境下，我們運用理論分析來評估行為相依戰略(behavior-dependent tactics)的有效性。接著，本研究提出一種新的框架(framework)以便將單議題戰略應用至多議題協商，並討論其優劣。最後，我們使用窮舉式模擬實驗來驗證各種不同戰略在雙議題協商時的效能。實驗結果除用以比較各種不同戰略在成交效用、成交比率以及公平等指標上之差異外，數據也顯示當雙方都使用同一種戰略時，比較容易取得雙贏的結果。此外，我們也發現:簡單的『以牙還牙』戰略(tit –for-tat tactic)就可獲得比其他複雜戰略更好的成交結果。 藉由上述介紹可知:本論文已針對電子化協商發展有效的支援方法，相信此研究成果必能進一步拓展電子商務架構下之協商交易，並有助於交易自動化的提昇。

關鍵詞: 協商戰略與策略(Negotiation Tactics and Strategies), 期望效用(Aspiration Utility), 協商結果預測(Prediction of Negotiation Outcome), 協商支援(Negotiation Support), 加總計分模型(Additive Scoring Model), 電子商務(E-commerce)

In the past few years, the footprints of e-commerce are ubiquitous and range from simple web shopping to complex virtual partnership construction through the Internet. Faced with diverse user groups, different trading models have been developed in e-commerce as well, including posted-price selling, auction and negotiation. Of the three models, negotiation is undoubtedly the most complicated one and is essential for a certain types of trading. However, at present, there is no evidence to show a widespread adoption of negotiation in e-commerce. It implies the techniques of assisting people to conduct negotiations in e-commerce need to be further investigated. To popularize the negotiation-type trading in e-commerce, software packages that help the users completing the preparation work and guiding the negotiation process are indispensable. For this purpose, computer-aided negotiation support systems can serve as effective tools to assist the negotiators. In particular, modern negotiation support systems benefit from the remarkable computation power of computers and possibly provide the negotiators with more comprehensive and sound suggestions than the human think-tank does. However, the success of applying negotiation support systems to e-commerce does not only depend on the efficiency of computers but also on the quality of support to which the support system really contribute. To provide effective negotiation support in e-commerce, in this thesis, several novel techniques are proposed for the following three important subjects: (1)deriving accurate and reasonable aspiration utilities for negotiators to help them establish precise goals before negotiation, (2)providing a reliable prediction of outcome of a negotiation to help the participants achieving a better settlement, (3) establishing a complete evaluation of the negotiation tactics to convince the users of relying the automated negotiation framework. To assist a negotiator in determining a precise aspiration utility before negotiation, an analytical approach based on the additive scoring model is proposed in our work. A series of simulation-based analyses were performed on two-party multi-issue negotiations. The results show that it is possible to determine a range for an aspiration utility without knowing the preferences of either side. Furthermore, if the preference of one side had been identified before negotiation, the results can be further restricted in a narrower range. To predict the outcome of two-party multi-issue automated negotiations, we first demonstrate the area of the settlement space is positively correlated to the possibility of settlement and the final joint utility, whereas it is negatively correlated to the negotiation period. Next, we show how to correlate the combination of an issue’s weight assignment for both sides with the area of the settlement space. Thus, the outcome of a negotiation can be estimated by calculating the area of the settlement space. Based on the negotiation model proposed by Faratin et al. (1998), this thesis also examines the performance of automated negotiation tactics and intends to provide concise suggestions for the users of automated negotiation systems. First, a theoretical analysis is provided for behavior-dependent tactics. Useful conclusions are obtained when single-issue negotiations are considered. Next, a new framework for applying tactics derived from single-issue negotiations to multi-issue negotiations is proposed and discussed. Finally, different from the previous work, exhaustive simulations based on two-issue negotiations are performed to evaluate the effectiveness of behavior-dependent and time-dependent tactics. The experimental results show that a relatively win-win settlement could be achieved when both sides use the same tactic. And, a simple tit-for-tat tactic would obtain better results than others. keywords: Negotiation Tactics and Strategies, Aspiration Utility, Prediction of Negotiation Outcome, Negotiation Support, Additive Scoring Model, E-commerce