Astudyoflinguisticmodelingfordecisionknowledgediscovery

[[abstract]]在快速變遷與動盪的商場環境中，「知識」已被視為公司企業與經濟體保持競爭優勢最具關鍵的資源，而知識管理也因此成為現代管理動中的主要議題，其涵蓋了企業內知識的建立、儲存、分享、以及再使用。然而在過去許多知識勘探的研究中所使用的方法論，仍以建立預測模式為主，而利用精確的數理模型來進行預測模式之建立對於知識的貢獻仍備受質疑，因此本研究即提出一套語意知識探勘法，提供一個改以「知識」為主題的資料探勘模式，以強化知識管理中的隱性知識移轉活動。本研究實際應用兩個案例進行知識法則之探勘，以驗證本研究所提出之方法。 從領域專家中挖掘其內隱知識在當前的知識管理上是一項十分有趣且極具挑戰的重要工作，本研究第一個應用個案即為財務報表分析專家決策過程的知識萃取，應用本研究所提出之模糊語意塑模法將專家決策之內隱知識予以萃取出來，並轉換為人類口語化之法則。實證資料來源將以國內數百家上市上櫃公司的財務報表資料為實驗對象，並商請一位資深會計師協助進行實驗，以驗證本研究所提出的方法能夠有效地將專家內隱知識外顯化，並得以修正及改善其決策知識。 除此之外，知識並非僅只存在於專家的思維中，知識亦存在於過去歷史現象之中，因此本研究的第二個應用案例即針對時間序列資料進行知識法則探勘，以不同之知識來源進行知識發掘的研究。本研究亦應用時間序列的資料處理法，將臺股大盤收盤資訊予以匯整並進行其特徵擷取，以配合模糊語意塑模法之使用。本研究再行配合語意塑模法，將上述所開採之模糊數學模式予以轉換為較廣為人接受的口語知識，以完成自時間序列資料中探勘隱藏於知識的目的。[[abstract]]In the ever-fast dynamically changed business environment, the key of remaining competitive advantage is the knowledge, the core capabilities of the organizational units in the global economic society. Thus knowledge management becomes the major issue of current management activities, which involves creating, innovation, repository, sharing, and re-use of mind’s sprit. There have been a lot of researches prove that IT can support knowledge management tasks, however, most of them are mainly focused on “knowledge prediction” as the research goal. In addition, empirical studies indicated that forecasting tasks are of limited contribution to knowledge management. Hence, to fit the goal for matching the contemporary knowledge society activities, this research propose an hybrid approach of fuzzy modeling and linguistic approximation, in order to enhance the tacit knowledge transfer for decision knowledge of a domain expert. Two empirical cases were conducted in this thesis for demonstrating the feasibility and capability of the hybrid approach. To elicit the tacit knowledge from domain experts is an exciting and challenging work. Therefore, the first case of this thesis focuses on financial statement knowledge discovery. To reach this goal, we propose an approach of linguistic modeling to transform the traditional fuzzy model into linguistic and human understandable knowledge instead of the mathematical equation in fuzzy model. Hundreds of Taiwan’s companies financial statement report are investigated , the experimental results show that financial decision knowledge extracted from this approach matches three main objectives in this research, which are accuracy, simplicity, and comprehensibility. The knowledge is not only captured in financial statement decision experts but also persisted in large sequence event databases, such as Taiwan Stock Exchange as in the second empirical case of this thesis. To reach the goal of time series knowledge discovery, this research collects and transforms the original sequences into discrete and window frame-based data set. Furthermore, the data set then is fed into the linguistic model to extract the interesting and hidden financial decision for investors. Those knowledge rules are all compatible with three major objectives in this research