本論文以電子化協商為基礎，發展兼具效率、準確性與實用性的對手喜好預測方法。首先，本研究結合Raiffa提出的加總計分模型與Faratin等人提出之策略模型，做為將對手協商行為量化的依據。根據此綜合模型，便可明確定義與對手喜好預測直接相關之待解問題。接下來，我們以基因演算法為基礎，發展預測演算法，以便同時顧及效率與準確性。由於求解的速度與準確性可藉由演化的代數來控制，使用者可在協商中依據臨場需求來設定解的品質。此外，前人研究中對於議題數、協商戰略係數與效用函數等參數的諸多限制，在此也被大幅放寬，將可有效提升預測結果之實用性。實驗結果顯示:本研究提出之方法能準確預測對手喜好，使用預測的一方也可獲得顯著的利益。根據統計，當雙方各議題出價區間相同時，使用預測的一方平均可取得15.3%的效用改善；當議題出價區間僅有80%重疊狀況下，也可增加14.3%的效用。值得注意的是:雖然使用預測的一方效用增加，但對手的收益並未因此而減損，顯示預測的一方有能力主導雙贏。 The thesis intends to develop effective methods to accurately predicts the opponent’s preferences for negotiation in e-commerce. First, the negotiation behavior of opponent is quantified by combining the scoring additive model and Faratin’s strategy model. According to the combined model, the problems of predicting the opponent’s preferences can be defined concisely. Then, we develop a prediction method based on genetic algorithm to compromise effectiveness and accuracy. Because the quality of solutions can be controlled by the number of evolution generations, users can easily adjust their goal according to the time limit. Moreover, the restriction on the number of issues, the coefficients of negotiation tactics and utility functions in previous works can be greatly relaxed by the proposed method. The experimental results reveal that our method can predict the opponent’s preference correctly. Besides, the results also show that adopting the prediction method can bring considerable profit. According to the statistics of experiment results, the predicting side can increase 15.3% utility when the offer zone of each issue is fully-overlapped; and the utility increase 14.3% when the offer zone of each issue is 80%-overlapped. In particular, although the utility of the predicting side is increased, the opponent’s profit doesn’t decrease for this reason. This result demonstrate that the predicting side can dominant the negotiation, meanwhile, maintain a win-win settlement.
