流程模型之建構與比較為流程挖掘技術中最為重要且基本。流程模型之建構演算法中,α-algorithm已應用在許多實際案例且獲得不錯的結果。然而,該演算法對於推論得之活動關係缺乏回饋確認之機制造成流程模型與實際路徑有所差異。此外,多數流程挖掘演算法多從活動關係角度發展模型,而非從流程模型之使用者角度發展之。是故,本研究將從使用者角度藉由修改及延伸α-algorithm,以系統化的方式處理「多重選擇」架構的複雜活動關係。再者,對於流程模型比較的問題,相關研究多著重在量化模型相似程度,而無法表現造成模型差異之因素,對此,本研究則提出兩個評估指標,Support 和 Confidence和一個模型比較表格。Support用以表示模型之間的相似程度,Confidence則是衡量模式之間相同的部分佔個別模型之比例,而模型比較表格則是提供造成流程模型差異之因素。本研究中將提出之方法實際應用於挖掘台灣某私立醫院幼兒科護理人員之工作流程,並比較該流程模型與由α-algorithm所挖掘之流程模型之差異。 在執行速率上,因本研究所提出之方案增添了回饋的動作,所以速率較α-algorithm及目前現有的相關演算法慢,但是所挖掘出之流程模型是較為貼近實際路徑記錄。而流程模型的比較,目前仍無一廣泛被接受的比較評估標準,而本研究所提出之方法優勢在於不僅能量化流程模型的相似程度,而且能提供使用者更多的資訊以供流程管理之用。
Process mining is getting more and more attention in many fields. Among the related important and essential techniques, building a model from event logs and process conformance are the most two to be used widely. For building a model, the popular α-algorithm has been used in many real cases and obtains good results. However, the difference between the activity relationships existing in flows/traces and that summarized byα-algorithm results from the lacks of error detection and decision feedback for these inferred activity relationships. Besides, most algorithms developed from the aspect of the activity relationship, not the users who will work following the process model. To overcome these two problems, a modified algorithm has been developed by extending the α-algorithm and a pattern “Option” is proposed to handle these complex activity relationships systemically. Moreover, for process conformance, the related methods of process conformance can quantify the difference or similarity between processes, while they can not show what causing the difference. To resolve the previous problem, this study provides two parameters, Support and Confidence, to quantify the difference/similarity between processes as well as offers the key factors causing the difference. “Support” is to evaluate the similarity between processes based on the activities and activity relationships; and “Confidence” is to measure the relation between processes based on the ratio of their identical parts to the compared process. The proposed method is applied to mine the process model of the staff nurses’ processes of Taiwan’s hospital pediatrics department and estimate the difference of this model and the model mined byα-algorithm. In the performance efficiency, though the proposed algorithm can not outperform other related algorithms, yet does capture more complete relationships between/among activities and does show more understandable process model. As to process conformance, there is still no accepted widely standard for measuring the difference between processes, therefore, this study can not provide the exact estimating value but find the low bound of the difference and the factors causing the difference with a simplified method.