本研究針對臺灣詐欺案件之類似案件檢索問題，提出一套結合法律專業知識的推薦系統。研究以檢察官撰寫之起訴書為資料來源，經解析與內容抽取後，建立結構化的案件資料庫，並以各案件「所犯法條」之組合作為驗證標準。在系統設計上，本研究導入法學專業知識，包括「客觀構成要件」與「事實型量刑因子」，並運用大型語言模型進行結構化資訊抽取；另以經驗法則建構「詐欺手法要素」作為比較基準。研究中比較多種推薦機制，包括語句組比較、語句分群與量刑因子選項對比，並引入 Distribution-Based Score Fusion（DBSF）方法以整合不同機制的推薦結果。同時，本研究亦採用語意嵌入模型之微調與基於 TF-IDF 的重新排序，兼顧語意與字面層次的相似性。在效能表現上，基於法學專業知識所建構之推薦機制（客觀構成要件與事實型量刑因子）整體優於以經驗法則（詐欺手法要素）為基礎之機制。最後，以起訴書所載「所犯法條」作為推薦案件命中判斷依據，在平均之precision、recall 與 F1-score 三項評估指標中，最佳實驗結果分別達到 85.22%、86.50% 與 83.96%。

This study proposes a recommendation system for retrieving similar fraud cases in Taiwan, grounded in legal expertise. Indictments drafted by prosecutors served as the primary data source; after parsing and information extraction, a structured case database was built, with the “combination of charged statutes” of each case serving as the evaluation benchmark. In system design, structured features derived from legal knowledge—namely actus reus and sentencing factors—were extracted using large language models, while custom fraud technique elements based on empirical rules were constructed for comparison. Multiple recommendation mechanisms were evaluated, including sentence group similarity, sentence clustering, and sentencing factor comparison, and a Distribution-Based Score Fusion method was introduced to integrate results across mechanisms. Performance enhancements further included fine-tuning of semantic embedding models and TF-IDF re-ranking to capture both semantic- and lexical-level similarity. Experimental results show that mechanisms grounded in legal knowledge (actus reus and sentencing factors) consistently outperform those based on empirical rules (fraud technique elements). Using applicable statutes in indictments as the ground truth for evaluation, the system achieved best results of 85.22% precision, 86.50% recall, and 83.96% F1-score.

[1] 內政部警政署, “內政部警政署 165 打詐儀表板,”[線上]. Available: https://165dashboard.tw/. [存取日期: 2 June 2025].
[2] 法務部,“法務部檢察機關公開書類查詢系統,”[線上]. Available: https://psue.moj.gov.tw/psiqs/. [存取日期: 2 June 2025].
[3] 司法院,“司法院裁判書系統,”[線上]. Available: https://judgment.judicial.gov.tw/FJUD/default.aspx. [存取日期: 2 June 2025].
[4] 司法院,“事實型量刑資訊系統,”[線上]. Available: https://intellisen.judicial.gov.tw/. [存取日期: 11 June 2025].
[5] 藍家樑, 中文訴訟文書檢索系統雛形實作, 國立政治大學資訊科學系, 2009.
[6] 謝淳達, 利用詞組檢索中文訴訟文書之研究, 臺北: 國立政治大學資訊科學系, 2005.
[7] 呂凱煜, 基於向量空間模型之智慧型文件搜尋系統開發-以台灣醫療糾紛判決書為例, 中山醫學大學醫學資訊學系, 2021.
[8] Paheli Bhattachary, Kripabandhu Ghosh, Arindam Pal, Saptarshi Ghosh, Methods for Computing Legal Document Similarity: A Comparative Study, 2022.
[9] 曹錫璋, 基於深度學習模型之判決書情境相似檢索技術, 臺中: 國立中興大學資訊科學與工程學系, 2021.
[10] Yinglong Ma, Peng Zhang, Jiangang Ma, An Efficient Approach to Learning Chinese Judgment Document Similarity Based on Knowledge Summarization, arXiv（Computing Research Repository, CoRR）, 2018.
[11] Zhilong Hong, Qifei Zhou, Rong Zhang, Weiping Li, Tong Mo,“Legal Feature Enhanced Semantic Matching Network for Similar Case Matching,”2020 International Joint Conference on Neural Networks (IJCNN 2020).
[12] 楊長諭, 運用句法規則於命名實體辨識之研究, 雲林: 國立雲林科技大學資訊管理系碩士班, 2013.
[13] Wan-Chen Lin, Tsung-Ting Kuo, Tung-Jia Chang, Chueh-An Yen, Chao-Ju Chen, and Shou-de Lin,“利用機器學習於中文法律文件之標記、案件分類及量刑預測,”International Journal of Computational Linguistics & Chinese Language Processing (IJCLCLP), 第十七卷, ROCLING XXIV精選論文特刊, 2012.
[14] Lu Gu, Wenjing Zhang, Yao Wang, Bo Li, Song Mao,“Named Entity Recognition in Judicial Field Based on BERT-BiLSTM-CRF Model,”2020 International Workshop on Electronic Communication and Artificial Intelligence (IWECAI), 2020.
[15] P.-H. Wu, C.-L. Liu, and W.-J. Li.,“An empirical evaluation of using ChatGPT to summarize disputes for recommending similar labor and employment cases in Chinese,” Proceedings of the Eighteenth International Workshop on Juris-Informatics, 2024.
[16] Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova,“BERT: Pre-training of deep bidirectional transformers for language understanding,”Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), 2019.
[17] Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakasiotis, Nikolaos Aletras, Ion Androutsopoulos,“LEGAL-BERT: The Muppets Straight Out of Law School,”Findings of the Association for Computational Linguistics: EMNLP 2020.
[18] Chaojun Xiao, Xueyu Hu, Zhiyuan Liu, Cunchao Tu, Maosong Sun,“Lawformer: A pre-trained language model for Chinese legal long documents,”AI Open, 第二冊, p.79–84, 2021.
[19] 張子軒, 法律案件推薦系統—以內容為基礎過濾, 國立陽明交通大學管理科學系所, 2023.
[20] Philip Chung and Akshay Swaminathan and Alex J. Goodell and Yeasul Kim and S. Momsen Reincke and Lichy Han and Ben Deverett and Mohammad Amin Sadeghi and Abdel-Badih Ariss and Marc Ghanem and David Seong and Andrew A. Lee and Caitlin E. Coombes and Brad B,“VeriFact: Verifying Facts in LLM-Generated Clinical Text with Electronic Health Records,”arXiv preprint arXiv:2501.16672, 1 2025.
[21] 王皇玉, 刑法總則, 第八版 編者, 新學林, 2022.
[22] Chao-Lin, Liu and Yi-Fan, Liu, “Some practical analyses of the judgment documents of labor litigations for social conflicts and similar cases,”CEUR Workshop Proceedings 3423: Proceedings of the Third International Workshop on Artificial Intelligence and Intelligent Assistance for Legal Professionals in the Digital Workplace (LegalAIIA 2023), ICAIL, Braga, Minho, Portugal, 2023.
[23] C. D. Manning, P. Raghavan, and H. Schütze, Introduction to Information Retrieval, Cambridge, U.K: Cambridge University Press, 2008.
[24] J. Richard Landis and Gary G. Koch,“The Measurement of Observer Agreement for Categorical Data,”Biometrics, 33, pp.159-174, 1977.