Clustering and Categorization of Brazilian Portuguese Legal Documents
Clustering and Categorization of Brazilian Portuguese Legal Documents
复制标题
巴西葡萄牙语法律文件的聚类和分类
DOI：
10.1007/978-3-642-28885-2_31
复制
发表时间：
2012
期刊：
影响因子：
--
通讯作者：
Vera Lúcia Strube de Lima
中科院分区：
文献类型：
--
作者：
Luis Otávio de Colla Furquim;Vera Lúcia Strube de Lima
本研究探讨了机器学习在电子审判案例法搜索中的应用。我们将案例法文档聚类，自动生成分类器的类。当用户将新文档上传到电子试验时，将使用这些类。我们选择了由Aggarwal，Gates和Yu创建的算法TClus，删除其文档/组丢弃功能并添加聚类划分功能。我们引入了一个新的范例“袋的条款和法律参考”，而不是“袋的话”，通过生成属性使用法律领域词库检测法律的条款和使用正则表达式检测法律参考。我们收集了一个判例法语料库。用相对硬度测量（RH）和硬度测量（RHO）对结果进行评价。结果进行了测试，与Wilcoxon的符号秩检验和计数的输赢检验，以确定其显着性。分类结果由人类专家进行评估。我们比较了真/假阳性对文档相似性的质心，集群大小，数量和类型的属性的质心和集群凝聚力。文章还讨论了属性生成及其对分类结果的影响。
This study explores the use of machine learning in case law search in electronic trials. We clustered case law documents, automatically generating classes to a categorizer. These classes are used when a user uploads new documents to an electronic trial. We selected the algorithm TClus, created by Aggarwal, Gates and Yu, removing its document/group discarding features and adding a cluster division feature. We introduced a new paradigm “bag of terms and law references” instead of “bag of words” by generating attributes using a law domain thesaurus to detect legal terms and using regular expressions to detect law references. We clustered a case law corpus. The results were evaluated with the Relative Hardness Measure (RH) and the-Measure (RHO). The results were tested both with Wilcoxon’s Signed-ranks Test and Count of Wins and Losses Test to determine their significance. The categorization results were evaluated by human specialists. We compared true/false positives against document similarity with the centroid, cluster size, quantity and type of the attributes in the centroids and cluster cohesion. The article also discusses attribute generation and its implications to the classification results.