图书情报工作 ›› 2023, Vol. 67 ›› Issue (2): 98-107.DOI: 10.13266/j.issn.0252-3116.2023.02.010

• 情报研究 • 上一篇    下一篇

面向局部可解释性机器学习的数据故事生成方法研究

肖纪文   

  1. 中国人民大学信息资源管理学院 北京 100872
  • 收稿日期:2022-04-19 修回日期:2022-10-21 出版日期:2023-01-20 发布日期:2023-02-09
  • 作者简介:肖纪文,硕士研究生,E-mail:xiaojiwen@ruc.edu.cn。
  • 基金资助:
    本文系国家自然科学基金项目“预测性分析结果的数据故事化描述方法及关键技术”(项目编号:72074214)研究成果之一。

Research on the Method of Data Story Generation for Local Interpretable Machine Learnin

Xiao Jiwen   

  1. School of Information Resource Management, Renmin University of China, Beijing 100872
  • Received:2022-04-19 Revised:2022-10-21 Online:2023-01-20 Published:2023-02-09

摘要: [目的/意义] 针对实践中数据故事应包含哪些内容、创作流程是什么等问题,提出一种数据故事生成方法,以期为数据故事的创作提供理论指导。[方法/过程] 在前人的研究基础上,基于数据科学、认知科学、自然语言处理和可解释性机器学习等理论,提出一种面向局部可解释性机器学习的数据故事生成方法,该方法对数据故事的生成步骤和创作方式进行详细的阐述和说明。同时对LIME算法的输出进行改进,使其更易理解。在此基础上对提出的数据故事化方法进行案例实现,以验证方法的可行性。[结果/结论] 提出的数据故事生成方法有助于丰富数据故事化研究的理论体系,同时为数据故事的生成研究和数据故事化工具的研发提供一定的启示。

关键词: 局部可解释性机器学习, 数据故事的生成, 数据故事化, 数据认知

Abstract: [Purpose/Significance] Data story has aroused extensive attention and application. Current research mainly focuses on theory such as the meaning or the model of data story, while there are lack of attention to practical problems such as what the data story should contain and what the creation process is. Therefore, this paper proposes a data story generation method so as to provide theoretical guidance for the creation of data stories.[Method/Process] Based on previous research, and according to theories of data science, cognitive science, natural language processing and interpretable machine learning, a method of data story generation for local interpretable machine learning was proposed and this method explained the generating steps of data story and creating methods in detail. At the same time, the output of the LIME algorithm has been improved to make it easier to understand. On this basis, a case implementation of the proposed data storytelling method was carried out to verify the feasibility of the method.[Result/Conclusion] The data story generation method proposed in this paper enriches the theory system of data storytelling research, and provides some enlightenment for the research on the generation of data stories and the development of data storytelling tools.

Key words: local interpretable machine learning, data story generation, data storytelling, data cognition

中图分类号: