透過研究ERP資料庫中資料的特徵，例如實體關聯模型中的實體屬性，來產生潛在的關鍵績效指標的方法在過去的研究中已被證實是可行的。指標的呈現，通常需要以不同的管理角度切割才能發揮反映營運現況的效果；因此使用者必須了解在目前資料庫留存資料可呈現的前提下，新產生的關鍵績效指標有何維度可使用。因此必須提出一套方法能夠自動化的產生潛在的關鍵績效指標及其多維度模型，此為本研究之動機。 先前對於自動化產生關鍵績效指標的研究，主要透過文字探勘的技術，比對系統中對於實體屬性的欄位敘述與資料倉儲(DW)系統中既有之指標運算域的敘述的相似度來進行實體屬性與運算域的連結，透過預先定義的公式結構來產生候選之關鍵績效指標。在過去的方法中，忽略英文中基於文法導致字綴不同的情況，此導致在字頻計算上的誤差，因此本研究應用字幹化演算法提升字頻計算的準確性。過去研究定義比對單位為實體中所有屬性之敘述，此導致產生過多無用的候選績效指標，本研究亦修正過去方法在定義比對單位上的錯誤，藉此降低產生的指標數達可進行人為篩選的數量。 候選績效指標由資料庫中之實體屬性所組成，實體屬性並非總是來自相同實體，在由來源系統資料模型產生多維度模型的研究中，並未討論定義多維度模型的事實表格(Fact table)時，若資料來自不同實體的處理方式；本研究透過最短路徑演算法，尋找連結組成候選績效指標之事實表格所需之實體間的路徑，透過結合(join)路徑上的實體以建構事實表格。此法能確保最少量的結合運算。並透過結合後之實體與修改後的實體關聯模型，套用目前已提出的多維度模型產生演算法產生候選指標之多維度模型。The method for generating potential Key Performance Index (KPI) by studying the characteristic of ERP database, such as Attribute of Entity in Entity-Relationship Model, is feasible, which is proven from previous research. In order to reflect the current operating status of enterprise, KPIs have to present in different managerial aspects. Therefore, user needs to know what dimension can be used to present the new-found KPIs under the existing data of database. As mentioned above, proposing an automatic method to generate the potential KPIs with its dimensional model is the motivation of this research. In previous research of automatic KPIs generating method, it compares the similarity of description of Entity-attribute in ERP database with the operand of existing KPI in DW system by text mining technique to make the linking of Entity-attribute and operand. And it uses the predefined formula structure to compose the KPI candidates. In the past method, it ignores the situation of affix difference which causes the error on counting term frequency. Hence, this research applies the stemming algorithm to improve the accuracy on the TFIDF weight calculation. Besides, because the past method defends entire entity as the compared unit, it causes too much useless KPI candidates and its definition is not reasonable. This research also corrects this problem on the definition of compared unit, and therefore reduces the quantity of new-found KPIs to the number that can be filtered by human. KPI candidates are composed by the Entity-attribute in database. The Entity-attributes used in those candidates are not always chosen from the same Entity. In the research of constructing Dimensional model form the data model of source system, it seldom discusses the problem which data is coming from different entities when defining the Fact table. This research uses the shortest path algorithm to find which path connects the used entities of KPI candidates. And the entities on the path are joined to construct the fact table of Dimensional model. This method ensures the minimum number of joining operation. Then we use the joined entity and Modified ER diagram to construct the Dimensional model of KPI candidates by the existing algorithm for dimension development
