计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (4): 112-119.DOI: 10.3778/j.issn.1002-8331.2108-0358

• 模式识别与人工智能 • 上一篇    下一篇

案件要素异构图的舆情新闻抽取式摘要

李刚,余正涛,黄于欣   

  1. 1.昆明理工大学 信息工程与自动化学院,昆明 650500
    2.昆明理工大学 云南省人工智能重点实验室,昆明 650500
  • 出版日期:2023-02-15 发布日期:2023-02-15

Extractive Summary for Public Opinion News via Case Elements Heterogeneous Graph

LI Gang, YU Zhengtao, HUANG Yuxin   

  1. 1.Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China
    2.Yunnan Key Laboratory of Artificial Intelligence, Kunming University of Science and Technology, Kunming 650500, China
  • Online:2023-02-15 Published:2023-02-15

摘要: 案件舆情摘要是指从与司法案件相关的舆情信息中抽取与案件相关的句子作为摘要。在案件舆情文本中通常包含如涉案人员、案发地点等案件要素,这些案件要素对于摘要生成有着重要的指导意义。因此,针对案件舆情文本的特点,提出一种基于案件要素异构图的抽取式摘要模型。首先通过基于图注意力机制融入案件要素的方法,构建一个由句子节点、词节点和案件要素节点组成的异构图,来捕捉句子间的关联关系,最后对句子进行分类,生成摘要。在基于百度百科构建的案件舆情数据集上进行实验,结果表明,模型相比基于注意力机制融入案件要素的方法在ROUGE-L上取得14.22个百分点的提升。

关键词: 案件舆情摘要, 案件要素, 图注意力机制, 异构图

Abstract: Case public opinion summary refers to extracting sentences related to the case as a summary from the public opinion information related to judicial cases. The public opinion text of the case usually contains case elements such as the person involved and the location of the case, these case elements have important guiding significance for abstract generation. Therefore, in view of the characteristics of the case public opinion text, an extractive summary model based on the heterogeneous graph of case elements is proposed. Firstly, a heterogeneous graph composed of sentence nodes, word nodes and case element nodes is constructed by integrating the case elements based on the graph attention mechanism, to capture the relationship between sentences. Finally, the sentences are classified to generate a summary. Experiment on the case public opinion data set based on Baidu Encyclopedia, the results show that the model in this paper achieves 14.22 percentage points improvement in ROUGE-L compared to the method of incorporating case element based on the attention mechanism.

Key words: case public opinion summary, case elements, graph attention mechanism, heterogeneous graph