由於國內詐騙案件發生數居高不下，政府要求警察機關將「打詐」列為治安重點項目，以維護民眾財產安全，提升民眾滿意度。2005年底，警察機關建置165反詐騙諮詢專線資訊系統（以下簡稱「165反詐騙系統」）作為受理民眾報案、諮詢跨機關單一窗口，並協調金融、電信、網路遊戲等民間資源整合，成為全國詐欺犯罪資料庫。無論受理報案之第一線員警或是偵辦詐欺案之司法警察，經常使用「165反詐騙系統」內之資訊。惟此系統在使用上仍有些不足之處：(1)勤務執行機構第一線員警受理詐欺案件除需制作調查筆錄，仍需在「165反詐騙系統」、「受理報案e化平臺」等不同警政系統平台輸入報案資料，曠日費時；(2)因「165反詐騙系統」內需點選輸入之資料欄位繁雜，以致經常發生第一線員警主觀上判斷錯誤、誤勾選項，間接使後續偵辦詐欺案之司法警察耗費許多時間在整理匯出之資料內容。本研究首先透過自然語言處理技術、推薦系統技術等機器學習技術解決以上之問題。在確認機器學習技術可應用在制作調查筆錄問句推薦、協助案件分類後，本研究更提出一項提高偵辦效率之偵查應用，並依據受理與偵辦模式設計一個基於自然語言處理與推薦系統協助警察偵辦詐騙案件之系統架構，作為未來警察機關改善建置之參考。;Due to the high number of domestic fraud cases, the government requires the police to list "Combating Fraud" as a key public security project to safeguard the safety of people′s property and improve public satisfaction. At the end of 2005, the police established 165 Anti-fraud consultation information system (hereinafter referred to as "165 Anti-fraud system") as a single window for accepting people′s reports, consulting across agencies, and coordinating the integration of private resources such as finance, telecommunications, and online games.This system became the national fraud database. The information in the "165 Anti-fraud System" is often used regardless of the first-line police officers who handle the report or the judicial police who investigate the fraud. However, there are still some shortcomings in the use of this system: (1) In addition to the need to produce survey transcripts, the first-line officers of the police service still need to make "165 Anti-fraud system" and "receive the report e-platform". The input of the report data of different policing system platforms will be time-comsumming; (2) Due to the complexity of the data fields for the "165 Anti-fraud System", the first-line police officers often make mistakes. These indirectly cause the judicial police who follow the fraud investigation to spend a lot of time sorting out the contents of the remitted information. This research first solves the above problems through machine learning techniques such as natural language processing technology and recommendation system technology. After confirming that machine learning technology can be applied in the production of survey transcripts to recommend and assist in the classification of cases, this study proposes an investigation application to improve the efficiency of investigation, and design a natural language processing and recommendation system to assist the police according to the acceptance and investigation mode. The system structure for investigating fraud cases serves as a reference for future police agencies to improve their construction.
