由ERP資料庫和BW系統自動化產生同義字之關鍵績效指標和多維度模型

Automatically generating Key Performance Indices candidates based on synonyms and dimensional model design from ERP database and BW system

陳雅鈞(Ya-Jyun Chen)

指導教授 ： 沈國基

國立中央大學/管理學院/工業管理研究所/碩士(2012年)

摘要

在先前的研究中，關鍵績效指標候選的產生，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，並且為績效指標的候選建立多維度模型。
在此研究中，我們透過將運算域的敘述以同義字的方式做關鍵字的展開，希望能夠比對到更多的實體性的欄位敘述來增加KPI候選的數量。另外，由於KPI內運算域的敘述為一個聚合的數值，所以沒有辦法透過字面上的比對來找到相似的實體屬性敘述；在本研究中,我們提供了解析聚合數值的演算法來為聚合運算域找到相關的實體屬性敘述。而在過濾關鍵績效指標的候選時，我們發現一部分的無意義指標的產生，是由於資料本身並沒有進行錯誤資料型態的過濾，在此研究中也會進行修正。最後，我們建立候選關鍵績效指標之多維度模型。為了提升多維度模型的管理意涵，在此研究中將會考慮到階層關係，增加可用來做為管理意涵的分析的維度表格。

關鍵字

多維度模型設計 ； 關鍵績效指標

並列摘要

In previous research, the authors treat operand in KPI formula as query word to find similar description of entity attribute to form KPI candidates. However, operand in KPI formula only contains 2-3 words, which are hard to perfectly perform a query so that it may causes worse mining results and also affect the number of KPI candidates generated. After generating KPI candidates, we discover that some aggregate value operand cannot find its mapping attributes. Furthermore, we discover that some of the dropped KPI candidates are due to the data type of operands in KPI candidates are not numeric value. When previous authors generate dimensional model, they only consider the entities that adjacent to fact table as dimension tables, ignoring other entities that connect to the adjacent entities, which may cause information lost.
In this research, in order to increase the number of KPI candidates we generated, we expand the operand words from existing KPI in Data Warehouse (DW) system based on its synonyms from lexical database, attaching these synonyms to operand words as query words. By text mining technique, we modified TFIDF to compare the similarity between description of entity-attributes and query words. Moreover, we also modified the predefined structure that used to generate KPI candidate by switching the operand set based on operator. Besides, in order to decrease the number of meaningless KPI candidates, we filter out those entity-attributes with uncountable data type. For those aggregated operands which cannot find mapping attributes, we proposed an algorithm to disaggregate it to find description of entity attribute. Moreover, eliminating entity-attribute of improper data type may also make TFIDF weighting procedure more precisely. Also, we improve the dimension model through merging the entities that connect to dimension table by their hierarchy.

並列關鍵字

dimensional model design ； Key performance index

參考文獻

38. 廖敏妏, 「由ERP資料庫和BW系統自動化產生流程導向之關鍵績效指標和多維度模型」，國立中央大學，碩士論文，民國99年。

36. 鐘國華，「利用企業資源規劃以及商業智慧資料產生銷售面績效評估指標」，國立中央大學，碩士論文，民國97年。

34. 吳佩蓉，「從 ERP交易資料發掘供應商績效評估指標並考慮查詢構面」，國立中央大學，碩士論文，民國95年。

33. Yutaka Matsuo& Mitsuru Ishizuka,2004, International Journal on Artificial Intelligence Tools,392-396

2. Boehnlein, M., & Ulbrich-vom Ende, A. (1999). Deriving initial data warehouse structures from the conceptual data models of the underlying operational information systems. Proceedings of the 2nd ACM International Workshop on Data Warehousing and OLAP, 15-21.