在製造業資訊系統日益普及的背景下，ERP系統不僅承載營運核心流程，也蘊含大量使用者操作行為的紀錄。本研究以提升ERP系統使用效率與使用者體驗為目標，針對系統日誌資料進行行為模式挖掘。研究動機源自實務觀察：使用者在系統中存在操作流程反覆、切換頻繁等現象，可能影響作業效率，亟需透過資料探勘技術進行分析與優化。 本研究採用Apriori（R語言實作）、FP-Growth與PrefixSpan等三種資料探勘演算法，建構分析模型，並以真實企業ERP操作日誌資料為實驗素材。資料經過前處理後，依使用者所屬地區與季度分群，進行關聯法則與序列模式探勘。Apriori與FP-Growth演算法在效能與規則完整性方面表現優異，能快速挖掘具代表性的高信賴度規則，有效擴展項目組合；PrefixSpan可掌握操作流程的時間順序，評估使用者是否依邏輯順序進行操作。 實驗結果發現，部分功能操作呈現高度共現與連續關係，如「工單短缺料查詢」與「銷售訂單查詢」、「銷售訂單釋出」與「出貨單查詢」等。亦發現用戶在介面上有頻繁切換或逆序操作的行為，顯示系統介面設計或流程規劃可進一步優化。整體而言，本研究成功整合三類探勘技術，提出具體改善建議，為ERP系統的流程設計與使用者經驗提供實證基礎。 ;With the growing prevalence of information systems in the manufacturing industry, ERP systems have become not only the backbone of enterprise operations but also a rich source of user interaction data. This study aims to enhance ERP system efficiency and user experience by mining behavioral patterns from system log data. The research is motivated by practical observations that users often exhibit repetitive operations and frequent task-switching, which may hinder workflow efficiency and highlight the need for data-driven analysis and optimization. To achieve this, the study applies three data mining algorithms—Apriori (implemented in R), FP-Growth, and PrefixSpan—to construct analytical models using real-world ERP operation log data. After preprocessing, the data is segmented by user region and quarterly periods for targeted pattern discovery. Apriori and FP-Growth both demonstrated strong performance in rule completeness and execution efficiency, capable of identifying representative and high-confidence rules while efficiently expanding item combinations. PrefixSpan was used to preserve temporal ordering and assess whether users performed actions in a logical sequence. Experimental results revealed high co-occurrence and sequential relationships between certain functions—for example, “shortage inquiry for work orders” and “sales order inquiry,” or “sales order release” followed by “shipping inquiry.” The analysis also uncovered frequent interface switching and reverse-order actions, indicating areas where interface design and process planning could be improved. Overall, this study successfully integrates three mining approaches to provide actionable recommendations, offering empirical support for optimizing ERP workflows and user experience.
