線上市集的競爭非常激烈且瞬息萬變，許多企業為了獲持續得競爭優勢，提供了多種的線上的服務與應用程式。然而，開發高品質的服務和應用程式來滿足使者的需求和喜好，供應商必須要了解和預測其使用者之交易行為。因為使用者使用不同的系統進行互動，而使得這些行為有所不同，例如，由不同供應商提供的線上購物、網路拍賣等系統。 本預測方法建立於內文資訊(可見的評分資料)或是交易數據(隱藏性的回應)中。然而，要求使用者提供內文資訊和明確的回應，同時更新關於個人的資訊與偏好，在實務上，這些非常困難。例如，在某些個案中， B2B的即時價格預測模型反而與線上拍賣結果相左，甚至內文資訊對於預測的貢獻更少。因此，本研究旨在開發有效的預測方法，此方法以使用者的交易資料(隱藏性的回應) 為基礎，當中包含了購買和議價資料，並提出了兩種預測方法。 第一種的預測方法，試圖從B2B 拍賣的收盤價和時間反向推估預測，而第二種方法預測使用者在電子商務中未來的偏好且提供建議項目。第一種方法的模擬是以 B2B線上拍賣市場的即時資料為基，時間超過兩年。模擬結果顯示，在觀察的第4次出價，此方法可以預測到收盤價格的84.6％，且準確率達71.9％。為了評估的第二個預測方法的效率和穩定性而進行了多次的實驗，資料來自於向3G服務供應商購買電影的網站。實驗結果顯示，第二種方法與傳統以內容為基礎的方法相比較，準確性平均提高了1.15倍。In very competitive and fast changing online marketplaces, many companies have been providing various amounts of online services and applications to gain sustainable competitive advantages. However, to develop high quality services and applications which meet users requirements and preferences, providers always need to know and predict their users transaction behavior. These behaviors vary from systems since users interact with different systems e.g. online shopping, online auction etc., that hosted by different providers. The prediction approaches can be developed based on whether contextual information/explict ratings or transactional data/implicit feedbacks. However, in practice, since contextual information and explicit feedbacks are obtained by asking users to provide and update information about their identity and interests, this information is difficult to collect and not always available. In some cases, for example, in real time closing price prediction model for business–to-business (B2B) reverse online auction, even the contextual information less contributes for prediction. Therefore, this study aims to develop efficient prediction approaches based on user’s transactional data/implicit feedback which are purchasing and bidding data and proposes two prediction approaches. The first prediction approach strives to predict closing price and duration of B2B reverse online auction, while second approach predicts user’s future preference for recommending items in m-commerce. The simulation of the first approach is based on real time auction data derived from a B2B online auction marketplace over a two-year period. Simulation results show that after observing the first 4 bids, this methodology can predict closing prices and duration with 84.6 % and 71.9 % accuracy, respectively. To assess the efficiency and stability of the second prediction approach, the several experiments are conducted on purchasing data collected from a movie website hosting by a 3G service provider. The experimental results show that the second approach improves the accuracy of traditional content-based approach by 1.15 times on average.
