用決策樹法則探討非對稱三人英國式拍賣的效率性質

Abstract

在給定估價函數的情形下，已經發掘出許多非對稱有效率英國式拍賣的充分條件。舉例來說，單次超越對手條件、均值超越條件與循序超越條件都是典型的案例。如何發現更多充分條件是一個研究上的主要目標。在本論文中，提出以決策樹演算法探勘出未知的充分條件。本論文建立了一個以電腦為基礎的探勘系統，它包含了拍賣資料產生器、關聯式資料庫、有效率拍賣的測試系統、決策樹的演算法。拍賣資料的產生是利用隨機產生的資料來建構估價函數。關聯式資料庫是用來儲存稍後統計上所需的資料。拍賣測試系統用來鑑定拍賣資料是否有效率。對通過測試資料的有效率拍賣而言，決策樹演算法是用來學習其可能的充分條件。這學習結果顯示出探勘有效率拍賣充分條件研究方向的可行性。

Many sufficient conditions of asymmetric efficient English auction have been developed from the given valuation functions. For example, pairwise single-crossing condition, average crossing condition, cyclical crossing condition are typical ones. Also how to find more sufficient conditions is the research goal. In this thesis, a decision tree algorithm is proposed to explore additional sufficient conditions. Moreover, a computer-based exploring system consisting of an auction data generator, a relational database, an efficient auction testing system, and a decision tree algorithm has been built. The auction data generator is used for generating random data for constructing valuation functions. The relational database is used for storing the generated data for statistic use lately. The auction testing system is used for identifying whether the auction data is efficient or not. Given the well-tested data of efficient auctions, the decision tree algorithm is used to learn the rules as their possible pre-conditions. The learning result is to show a possible research direction for exploring more sufficient conditions of efficient auction.