基于法院判决文书的法律知识图谱构建和补全

Construction and Completion of Legal Knowledge Graph Based on Court Judgment Documents

由于法律领域知识图谱专业性强、结构复杂,而现有的关系抽取方法因各个领域的需求和术语不同,无法适用于法律领域知识图谱的构建和补全。首先,提出了基于StanfordNLP关系抽取机制的法律知识图谱构建方法;然后,构建基于设置谓语导向词的深度学习模型对法律知识图谱进行补全;最后,选用典型案例(伪卡盗刷判决书)作为文本对象验证模型的可行性。与其他知识图谱补全模型相比,本模型的准确率达到95%以上。基于谓语导向词的深度学习模型综合了自动构建和人工参与,提高了关系抽取的准确率和补全的效率,能最大程度挖掘判决书文本中的深层隐式关系,更好地发挥判决书文本的应用技术。

Due to the specialty and complex structure of the legal-domain-knowledge-graph,the existing relationship extraction methods were not suitable for the construction and completion of the legal-domainknowledge-graph because of the various needs of different domains.A legal-knowledge-graph-construction method based on StanfordNLP relation extraction mechanism was proposed.Then,a deep learning model based on predicate-oriented words was proposed to complete the legal knowledge graph.At last,a typical case(the judgment of counterfeiting card theft)was selected as the text object to verify the feasibility of the model.Compared with other knowledge graph completion models,the accuracy of this model exceeded 95%.The deep learning model based on predicate oriented words integrated automatic construction and human participation,which improved the accuracy of relation extraction and the efficiency of completion.This method could excavate the deep implicit relationship in the text of the judgment to the greatest extent and achieved better application of judgment text technology.