情报科学 ›› 2025, Vol. 43 ›› Issue (3): 176-184.

• 博士论坛 • 上一篇    下一篇

基于可解释性理论的数据故事化解释效果评价研究

  

  • 出版日期:2025-03-05 发布日期:2025-05-27

  • Online:2025-03-05 Published:2025-05-27

摘要: 【目的/意义】探索基于可解释性理论的数据故事化解释效果评价方法对于拓展可解释性方法、推动故事化 应用具有重要意义。【方法/过程】在述评可解释性理论、数据故事化的解释作用及解释效果评价等内容的基础上, 构建了一个结合可解释性结果的数据故事化解释模型及公式化表达和评价指标体系。本研究通过实证分析,采用 问卷调查法的半定量分析与基于指标参数的定量分析方法,对所提出的评价方案进行验证,明确了数据故事化在 增强结果可解释性中的作用。【结果/结论】将可解释性结果以数据故事的形式呈现,能显著提高用户的理解效果和 接受度。本研究不仅丰富了可解释性理论和数据故事化领域的研究,也为实践中评价模型解释效果提供了有价值 的指导方案。【创新/局限】数据故事化的解释效果评价为衡量可解释性、可理解性等指标提供了新的视角与方案, 但对于评价模型的改进与优化策略有待深入研究。

Abstract: 【Purpose/significance】It is of great significance to explore the evaluation method of data storytelling interpretation effect based on interpretability theory to expand interpretability methods and promote the application of data storytelling.【Method/process】 On the basis of reviewing the interpretability theory, the explanatory role of data storytelling and the evaluation of interpretation effect, a data storytelling interpretation model and a formulaic expression and evaluation index system combined with interpretability results were constructed. Through empirical analysis, semi-quantitative analysis based on questionnaire survey method and quantitative analysis method based on index parameters, this study validates the proposed evaluation scheme, and clarifies the role of data storytell⁃ ing in enhancing the interpretability of results.【Result/conclusion】The results show that presenting the interpretability results in the form of data stories can significantly improve the user′s comprehension and acceptance. This study not only enriches the research in the field of interpretability theory and data storytelling, but also provides valuable guidance for evaluating the effectiveness of model in⁃ terpretation in practice.【Innovation/limitation】The evaluation of the data storytelling interpretation effect provides a new perspective and scheme for measuring indicators such as interpretability and comprehensibility, but the improvement and optimization strategies of the evaluation model need to be further studied.