Scholarlink

Legal documents tend to be lengthy, requiring significant effort to analyze them. Additionally, they contain a large amount of legal terminology, making it unsuitable to extract conclusions from them using traditional models. This article presents different machine learning models to perform the task of rhetorical role prediction in Hindu judicial rulings as part of SemEval-2023 Task 6: Legal Text Understanding, Subtask A. Experiments were conducted with tree-based and probabilistic models, such as Random Forest, J48, Support Vector Machine (SVM), Logistic Regression, and Naive Bayes, to predict Rhetorical Roles (RR) in legal document texts. Based on the results obtained, experiments and analyses were carried out using the aforementioned models and considering some combinations of them. The best result, evaluated with F1, was achieved by the Random Forest model, scoring 0.5258 (52.58%) using N-grams (trigrams).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Rhetorical Role Prediction in Legal Documents: A Comparison of Machine Learning Algorithms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hansel Lázaro Valdés Pérez,
  • Gabriela Giovana Pérez López,
  • Darnes Vilariño Ayala,
  • Beatriz Beltran Martinez,
  • David Pinto Avendaño

摘要

Legal documents tend to be lengthy, requiring significant effort to analyze them. Additionally, they contain a large amount of legal terminology, making it unsuitable to extract conclusions from them using traditional models. This article presents different machine learning models to perform the task of rhetorical role prediction in Hindu judicial rulings as part of SemEval-2023 Task 6: Legal Text Understanding, Subtask A. Experiments were conducted with tree-based and probabilistic models, such as Random Forest, J48, Support Vector Machine (SVM), Logistic Regression, and Naive Bayes, to predict Rhetorical Roles (RR) in legal document texts. Based on the results obtained, experiments and analyses were carried out using the aforementioned models and considering some combinations of them. The best result, evaluated with F1, was achieved by the Random Forest model, scoring 0.5258 (52.58%) using N-grams (trigrams).