一個以SCA與ESB為基礎的服務導向專家系統─以財務預警系統為例

Abstract

近年來網路科技發達，網路上充斥著各式各樣的資訊，其資料儲存位置越來越分散，格式也越分歧，因此資料整合和提供整合性加值服務漸漸成為一個重要的議題。因此本論文以服務元件(Service Component Architecture, SCA)為架構建置一個服務導向專家系統(Service-Oriented Expert System, SOES)，將不同平台上的服務變成一個可在網路上通用的服務元件(Service Component)，並結合企業服務匯流排(Enterprise Service Bus, ESB)串連全部的服務元件提供多樣化應用服務，同時利用專家規則結合資料探勘(Data Mining)的方式進行資料分類和預測。本論文將此系統架構應用在財務預警系統上，資料來源是以2006-2008年份台灣經濟新報資料的電子工業資料為主。模擬實驗先以2006-2007年的資料進行訓練，結合專家規則和決策樹方法預測2008年有財務危機的公司。透過模擬結果分析，我們發現我們所提出的系統可解決各公司財務服務間異質性問題，快速地推論潛在的財務危機公司，並簡化各公司定期上傳財務資料至台灣證券交易所之流程和人力整理成本。比較Altman多變量區別分析法與本論文所提出的方法，在評估規則正確率時，分別為73.24%和98.31%。故本論文所提出的方法改進34.2%正確率，因此專家規則結合決策樹之推論方法是能提供更高的準確度。

With the rapid growth of the Internet, there is a variety of information which is stored and distributed in the difference web sites on the Internet. So data collection and integration for providing value-add services are important issues. Therefore, we propose the Service-Oriented Expert System (SOES) based on Service Component Architecture (SCA) which can build the service components for each web services, combine the Enterprise Service Bus (ESB) for integrating these service components from heterogeneous platforms, and use both the expert rules and data mining technique to provide the data classification and prediction. In this paper, we apply the SOES to analyze the finance data which is derived from electronics in Taiwan Economic Journal (TEJ) during 2006 to 2008 for discovering the financial crisis companies. In experiments, the training data use the finance information gathered between 2006 and 2007, and then SOES uses expert rules and decision tree to explore the financial crisis companies in 2008. Through this architecture simulation, we can solve the problems by integrating heterogeneous services, inferring the potential financial crisis company rapidly, simplifying the data uploading procedure from company to the Taiwan Stock Exchange Corporation (TWSE), and saving the cost for getting and integrating financial data. Compared Altman’s multivariate discriminate analysis method with our research method are 73.24% and 98.31%, and the estimation of rules accuracy our approach improve the 34.2%. Therefore, using expert rules and decision tree to find the financial crisis company is higher performance.