韩国性犯罪案件法律文本分类：从传统机器学习到LLM及XAI洞察
原标题：Legal text classification in Korean sexual offense cases: from traditional machine learning to large language models with XAI insights
AI 摘要
该研究基于韩国性犯罪判例的十个法律类别，系统评估了从传统机器学习到大型语言模型的法律文本分类方法。实验结果显示，在韩国法律数据上微调的小规模模型 KLUE-BERT 准确率达 99.3%，优于 GPT-3.5、GPT-4.0 等大型通用模型及传统机器学习方法，表明领域适配与微调比模型规模更关键。研究还利用可解释 AI（XAI）分析模型预测与误分类案例，并借助 KICS 数据检验泛化能力，发现模型难以理解法律文本中的隐含语境线索，强调法律 AI 需兼顾高性能与可解释性。
正文节选
[Page 1] Artificial Intelligence and Law https://doi.org/10.1007/s10506-025-09454-w ORIGINAL RESEARCH Legal text classification in Korean sexual offense cases: from traditional machine learning to large language models with XAI insights Jeongmin Lee1,2 Accepted: 24 March 2025 © The Author(s) 2025 Abstract The advancement of natural language processing (NLP) has expanded the applica- tion of AI-based text classification in the legal domain. However, accurately clas- sifying legal documen