運用文本挖掘技術分析政府採購法之圍標相關判決案件

Analyzing Bid-Rigging Related Judicial Cases of Government Procurement Law Using Text Mining Techniques

過往研究多以人工標記判決書資料，並以統計方法來歸納其特徵，故本研究旨在從非結構化數據中萃取關鍵信息，藉由正規表達式進行特徵工程，並找出頻繁出現之項目集，助審計人員更高效準確地選取高風險的政府採購案件。本研究選擇Apriori與LDA主題分析在資料集中找出違反政府採購法之廠商、採購標案、招標機關之特徵與其頻繁組合。此方法可標示具圍標案件特徵的採購案，並歸納出涉及違反政府採購案常見特徵之頻繁項集，研究結果發現廠商為有限公司，且登記資本額介於100萬元至1,000萬元，參與最低標金額者出現機率為0.62，是為較頻繁項及；土木工程承包、建築業務和地方政府合作者距離相近，此些發現在有限審計人力下可成為挑選投標文件的參考之一。

Most previous studies have manually labeled judgment data and used statistical methods to summarize their characteristics. Therefore, this study aims to extract key information from unstructured data, perform feature engineering through regular expressions, and find frequently occurring item sets. , helping auditors select high-risk government procurement cases more efficiently and accurately. This study uses Apriori and LDA topic analysis to identify the characteristics and frequent combinations of manufacturers, procurement bids, and bidding agencies that violate government procurement laws in the data set. This method can mark procurement cases with the characteristics of bid-rigging cases and summarize frequent items involving violations of common characteristics of government procurement cases. The research results found that the manufacturer is a limited company with a registered capital between NT$1 million and NT$10 million. The probability of occurrence of those who participate in the lowest bid amount is 0.62, which is a relatively frequent project; civil engineering contracting, construction business and local government partners are close to each other. These findings can become one of the references for selecting bid documents under limited audit manpower.