【作者】 王永军；

【导师】 高静；

【作者基本信息】 内蒙古农业大学 ， 计算机应用技术， 2021， 硕士

【摘要】 随着大数据与人工智能时代的到来,将互联网技术、人工智能技术、大数据思维等信息化手段运用到纪检监察各项工作流程中,已成为主要方向。针对纪检监察工作领域的案件数量多、内容复杂、数据整理工作量大、定性量纪与定罪量刑工作效率低下的问题,相似案例智能推荐是缓解这一情况的一项重要措施。本研究基于知识图谱技术与人工智能技术,提出了一种融合纪检监察法律知识图谱特征表示的相似案例智能推荐算法。相比于传统的基于关键词的TF-IDF机器学习方法,提出的算法在F值上有较好的提升,可以为纪检监察日常工作带来便利,促进智慧纪检监察的建设。本研究具体工作如下:(1)纪检监察领域法律法规知识图谱的构建及搜索引擎设计与实现。近年来知识图谱在文本表示、推理方面的强大能力受到越来越多的关注。针对目前将知识图谱技术应用到纪检监察领域的研究还不是很多,本文的第一个研究内容是构建纪检监察法律法规知识图谱,通过知识图谱对法律条例中的法律实体与实体之间的关系进行语义表示,并使用Django网站开发技术与Echarts.js可视化库开发一个知识图谱搜索引擎,方便相关人员对知识图谱进行检索与分析。(2)纪检监察相似案例推荐数据集的构建。在纪检监察领域,目前还没有一个相对标准的违纪案例数据集供研究使用。本文搜集了近年来全国各省的3500多条违纪案例数据,并对获取的源数据进行过滤、清洗、标注,从而构建了一个纪检监察领域的违纪案例数据集供相似案例推荐研究使用。(3)融合法律法规知识图谱中领域知识的相似案例推荐算法的研究。本研究将相似案例推荐任务看成是多标签文本分类问题,在构建的法律法规知识图谱和违纪案例数据集的基础上,针对传统推荐模型未能充分考虑与违纪案例相对应法律知识的不足,本研究提出了一种融合了法律法规知识图谱的领域知识的相似案例推荐算法。首先对法律法规知识图谱中的相关实体与关系进行检索与定位,并将对应的法规内容提取并注入到案例推荐任务的文本特征表示中。使用几种机器学习方法与深度学习方法来进行对比实验,实验结果表明,本文提出的算法提高了相似案例推荐任务的准确率与可解释性。更多还原

【Abstract】 With the advent of the era of big data and artificial intelligence,the application of information technology such as Internet technology,artificial intelligence technology and big data thinking to the various work processes of discipline inspection and supervision has become the main direction.In view of work problems in the field of disciplinary and supervision such as the large number of cases,the complexity of the content of the documents,the large workload of data compilation,and the low efficiency of conviction and sentencing,intelligent recommendation of similar cases is an important measure to alleviate this situation.Based on knowledge graph technology and artificial intelligence technology,this research proposes an intelligent recommendation algorithm for similar cases that integrates the feature representation of the legal knowledge graph of discipline inspection and supervision.Compared with the traditional machine learning method keyword-based such as TF-IDF,the proposed algorithm has a better F-score,which can bring convenience to the daily work of discipline inspection and supervision and promote the construction of intelligent discipline inspection and supervision.The specific work of this research is as follows:(1)The construction of a knowledge graph of laws and regulations in the field of discipline inspection and supervision and the design and implementation of the search engine.In recent years,the powerful capabilities of knowledge graphs in text representation and reasoning have received more and more attention.There are not many studies on the application of knowledge graph technology to the field of discipline inspection and supervision.The first research content of this article is to construct a knowledge graph of laws and regulations in the field of discipline inspection and supervision and use the knowledge graph to semantically express the relationships between legal entities and entities in laws and regulations.Use Django——a website development technology and Echarts.js——a visualization library to develop a knowledge graph search engine,it is convenient for relevant personnel to search and analyze the knowledge graph.(2)The construction of dataset about similar case recommendation in discipline inspection and supervision.In the field of discipline inspection and supervision,there is currently no relatively standard dataset of violation cases for research and use.This research has collected more than 3,500 disciplinary case data from various provinces across the country in recent years,and the obtained source data have been filtered,cleaned,and annotated.And construct a disciplinary case dataset in the field of disciplinary inspection and supervision for use in recommendation of similar cases.(3)Research on the recommendation algorithm of similar cases that integrates the knowledge of the knowledge graph of laws and regulations.This study regards the similar case recommendation task as a multi-label text classification problem.Aiming at the traditional recommendation model that fails to fully consider the legal knowledge corresponding to disciplinary cases,based on the construction of the legal and regulatory knowledge graph and the case dataset,this research proposes a similar case recommendation algorithm that integrates knowledge of laws and regulations knowledge graphs.Firstly,search and locate related entities and relationships in the knowledge graph of laws and regulations,then extract and inject the corresponding legal contents into the text feature representation of the case recommendation task.Through comparative experiments of several machine learning methods and deep learning methods,the experimental results show that the algorithm proposed in this research improves the accuracy and interpretability of similar case recommendation tasks.更多还原