| 摘要: | 商業環境競爭之激烈與日劇增，許多公司在資訊系統的運用範疇也隨之擴大；從支援日常作業活動到支援高階主管決策之系統。研究發現，決策支援系統已成為組織提升競爭優勢與協助高階主管制訂決策所不可或缺的元素。而優良的決策支援系統需建構在能儲存、整合、分析與提供報表功能之系統上，商業智慧系統與相關的應用工具因運而生。 近年來，商業智慧系統受到學術界與產業界之重視，由探討相關議題的大量研究與報導可見一斑。許多IT專家與學者投入大量心力研究商業智慧系統之模型建置，採納來源資料以臻最大效率為大家所認同。其中，個體關聯圖（ＥＲ圖）一直為資訊業用來表達作業面資料之最佳工具，許多研究已提出不同的方法論，利用企業既有的ER圖作為基礎，快速建置商業智慧系統所需的模型，即多維度模型（multidimensional model）或稱星狀綱要 (star schema)。然而，文獻整理後我們發現，目前並未有任何研究提供方法可以完整地、有系統地將一個完整的ER圖轉換成商業智慧系統所需之具有階層架構的雪花狀多維度模型。因此，本研究的第一部份提出一個演算法來完成此目標，該研究有兩點特色：(1) 維持既有的資料顆粒度（granularity）(2) 以最小關係來完成多維度模型資料之合併。前者確立本研究所提方法之嚴謹度，確保資料顆粒度在資料轉換後不被破壞。對於有可能破壞資料顆粒度的實體關係，可在兩實體間建置一中介表格 (bridge table)，中介表格內置放權重因子（weight factors）做為調整資料顆粒度的橋樑。後者則強調若轉換過程有多種路徑選擇時，最小路徑將會是我們的選擇的路徑，以讓轉換後的多維度模型可以以最少的結合作業完成資料合併，使其成為較有效率的模型。最後，我們利用本研究所提的方法建置一個雛形系統，除作為展示外，並作為驗證所提方法論之可行性。 此外，商業智慧系統存放大量企業資料，無論在使用對象與功能上均具有獨有的特色，不同於一般資訊系統。亦有許多學者探討影響企業建置資料倉儲或商業智慧系統之關鍵因素；如商業智慧系統是否可以即時、以易懂方式提供高品質的資訊；商業智慧系統所涵蓋之資訊或設計是否納入企業的核心價值，方能提供真正的有競爭商業資訊給管理階層；商業智慧系統是否將內部與外部環境因素納入考量以及高階主管的支持與否等都是影響最後企業建置商業智慧系統之重要因素。另一方面，亦有研究指出，大部分的企業即使已經建置商業智慧系統，但其利用率卻很低。加上商業智慧系統的建置需投入大筆資金與時間人力，系統之投資報酬率又不易評估，這些均影響企業是否建置商業智慧系統之重要因素。 改善商業智慧系統效率問題因而受到學者之重視，然大部分的研究屬於技術導向，如在模型上的精進、查詢運算的強化、終端工具的開發、或是以更有效率查詢的資料報表呈現介面。並未有研究考慮使用者的特質或資料特性，提升有意義的資訊來提升使用者使用商業智慧系統之動機。因此，本研究的第二部份將應用資料礦採（data mining）技術，結合商業智慧系統所記錄的使用者查詢日誌，建置商業智慧系統的推薦系統。本研究目標是，提供使用者在制訂決策過程，有更多可以運用商業智慧系統的方向，俾使查詢多元化，除提升系統利用率外，進而獲取更多決策所需資訊。 The scale of Information Systems applications is expanding rapidly as the business environment becomes increasingly complex; for example, apart from supporting daily operations, systems are becoming more specialized to support management in decision-making. It has been found that decision support systems (DSS) are critical to helping organizations gain a competitive advantage and in helping managers make decisions. Excellent decision support systems have been constructed to store, integrate, and analyze data, and to provide reporting functions. As a result, business intelligence systems and related application tools are emerging. Many IT practitioners and researchers advocate that, to achieve maximum efficiency, data warehouse models should incorporate the sources of the data. As the source data is probably derived from systems designed with ER diagrams, a great deal of research has been devoted to the design of methodologies for building multidimensional models based on source ER diagrams. However, to the best of our knowledge, no algorithm has been proposed that can systematically translate an entire ER diagram into a multidimensional model with hierarchical snowflake structures. To fill this research gap, in the first part of this study, we propose an algorithm that achieves the above goal because it incorporates two features, namely, grain preservation and the minimal distance from each dimension table to the fact table. The grain preservation feature guarantees that the translated multidimensional model will maintain cohesive granularity among the entities. Meanwhile, the minimal distance feature guarantees that if an entity can be connected to the fact table in the multidimensional model by more than one path, the path with the smallest number of hops will always be chosen. The first feature is derived by (1) translating ambiguous relationships between entities into weighting factors stored in bridge tables, and (2) enhancing fact tables with unique primary keys. The second feature results from including a revised shortest path algorithm in the translating algorithm, with the distance being calculated as the number of relationships required between entities. A prototype system based on the methodology is also developed, and snapshots of the screens used for the system's execution are presented. In addition, data warehouses contain vast amounts of data with unique characteristics; hence, they are different from other information systems used in business enterprises. Several IT studies have investigated the critical success factors of data warehouses, including the availability of high quality information that is well organized, presented in a timely manner, and easily understood. A data warehouse has to be based on a core competence so that it can provide vital business intelligence to assist top managers in the decision-making process. Moreover, it should incorporate external and internal knowledge acquired over time and adapt such knowledge to current corporate conditions to support top management. Meanwhile, some studies have observed that there is low utilization of data warehouses in most organizations, and the amount of money that must be invested is much higher than for other information systems. Both low utilization and high investment costs can prevent an organization building a data warehouse system. Although several studies have addressed the efficiency improvements derived from business intelligence systems, most of the studies are technique oriented. They focus on the data models, query expressions, client tools, or materialized views for efficient query processing. None of them consider users’ characteristics or the data’s features to increase user interest and facilitate use of the systems. Therefore, in the second part of this study, we apply data mining techniques to business intelligence query logs to develop a recommendation mechanism. Our objective is to help users obtain more information by increasing their usage of business intelligence systems. |
