應用文本探勘技術之會計分錄測試方法

大數據時代來臨，審計作業可結合新興的資訊技術，以分析快速增長的大量資料。如此在原有的審計目標下，可有效加強審計執行程序的效果與效率。長期以來由於會計分錄摘要內容均以簡短文字呈現，故不易使用傳統資料分析工具予以分析；因此本研究應用文本探勘技術，分析會計分錄摘要的非結構內容，辨別及評估其風險，以協助審計人員更有效地找尋異常之會計傳票。本研究設計採用設計科學研究方法，將文本探勘技術結合資源－事件－人員模型（Resource-Event-Agent Model, REA Model），建置會計分錄風險評估系統雛型，以探索文本探勘技術應用於分析會計傳票風險之可行性。其分析程序主要針對單一會計科目的相關分錄摘要內容，應用REA模型進行分類、歸納及風險評分；以提供不同風險層級之篩選條件，利於會計師找尋可能誤記或蓄意隱匿之會計分錄。後續並將所設計的系統雛型交由會審專家評估其有效性及可行性，並提出回饋意見與建議。

關鍵字

會計分錄測試；風險評估；文本探勘；特徵選取

並列摘要

With the advent of big data, auditing work has combined with emerging information technologies to analyze the increasing volume of data. Without altering original audit objectives, applying new technologies can strengthen the effectiveness and efficiency of the audit process. Since the description of journal entries has been for long time recorded in short text, it is not easy to assess the risk of misrepresentations or intentional omission in journal entries by means of traditional data-analysis techniques. This study applies text mining techniques to analyze unstructured texts to find suspected abnormal journal entries. Based on the design science research methodology (DSRM) as a research method, we combine text mining techniques and the resource-event-agent (REA) model, to develop a system prototype for risk assessment of journal entries. Its purpose is to explore the feasibility of text mining techniques applying in the analysis of journal entry descriptions. The analysis procedures focus on the descriptions of each account in the journal entries with an application of REA model to classification, generalization, and risk scoring. These provide the filtering condition for different risk levels and assist auditors for locating journal entries caused by unintentional errors or deliberate concealment. The system prototype has further been offered to experienced audit experts to evaluate its effectiveness and feasibility and to provide feedbacks and suggestions.