詐欺案件法律文件生成與驗證：基於構成要件分析與判決書風格的雙階段應用框架

Abstract

本論文提出了一種基於刑法犯罪三階理論所構建的法律本體──刑法三階理論數學模型，並利用模型中的構成要件階層進行自動量化分析特定罪名。近年來，生成式人工智慧的興起，使得自動文本生成技術容易取得及使用，影響了諸多的產業，法學界及法律實務界也感受到了這股勢不可擋的潮流。但如何以自動化的方式評估生成式人工智慧所生成的法律文本內容，是一項具深具挑戰性的任務。本文依據刑法三階理論，以統一塑模語言（Unified Modeling Language, UML）來建構一個標準化可機械性操作的法律數學模型。由於該模型標準化的高度可操作性，並以刑法三階理論為基礎建立，是以該數學模型除了具備可解釋性外，亦容易轉換為電腦程式，使我們能夠以自動化的方式客觀評估生成式人工智慧所生成之法律文本，同時，應用在具體案件中，能幫助使用者基於刑法三階理論進行初步推理與評估生成式人工智慧所生成的文本完備性以及寫作風格，從而協助法律實務工作者具體修正生成文本內容的方向。本文所提出的模型不僅能提供法律資訊系統在推理能力的可解釋性，也為法律文件處理的自動化帶來新的應用方向，有望在未來實現更高效與準確的法律推理輔助工具。

This paper proposes a legal ontology based on the three-stage theory of criminal law, specifically its corresponding mathematical model, and utilizes the hierarchical structure of constituent elements within the model to quantify and analyze specific crimes automatically. In recent years, the rise of generative artificial intelligence has made text generation technology readily accessible and widely applicable, impacting numerous industries. The legal community and legal practice have also been affected by this unstoppable trend. However, evaluating the content of legal texts generated by generative artificial intelligence in an automated manner remains a highly challenging task. To address this, the paper employs the Unified Modeling Language (UML) to construct a standardized and mechanically operable mathematical model of law based on the three-stage theory of criminal law. Due to its high operability, the standardized model is not only interpretable but also easily translatable into computer programs, allowing for the objective and automated evaluation of legal texts generated by generative AI. When applied to specific cases, the model can assist users in making preliminary inferences grounded in the three-stage theory of criminal law and in assessing the completeness and writing style of AI-generated texts. This, in turn, helps legal practitioners refine and correct the content direction more precisely. The proposed model not only clarifies the reasoning capabilities of legal information systems but also introduces new applications for automating legal document processing, with the potential to lead to more efficient and accurate legal reasoning tools in the future.