摘要
[目的/意义]利用可解释性技术与故事化方法研究网络欺凌检测有助于识别欺凌内容、参与网络言论治理以及净化网络生态。[方法/过程]详细分析LIME解释算法的选择依据与文本解释原理,并提出LIME算法在网络欺凌检测模型中的“钻石结构式”应用流程,进一步构建了“数据层—模型层—解释层—叙事层”的故事化呈现框架,最后通过实验验证了该框架的有效性。[结果/结论]可解释性技术辅助网络欺凌检测模型有利于提高模型应用价值与可信度,权衡模型准确性与可解释性二者的关系,而基于数据分析及解释结果的故事化呈现方法为网络信息平台数据治理提供可信、可靠、可理解依据。
[Purpose/Significance]Using interpretability technology and storytelling methods to study cyberbullying detection can help identify bullying content,participate in online public opinion governance,and purify the online ecology.[Method/Process]The study analyzed the selection basis of LIME interpretability algorithm and text interpretability theory in detail and proposed the“Diamond”application process of LIME algorithm in cyberbullying detection model,further constructed a storytelling presentation framework of“data layer-model layer-explanation layer-narrative layer”,and final-ly verified the effectiveness of the framework through experiments.[Result/Conclusion]The interpretability technology for cyberbullying detection model can improve the application value and the model credibility,balance the relationship between model accuracy and interpretability,and the storytelling presentation method based on data analysis and interpretation re-sults can provide a credible,reliable and understandable basis for network information platform data governance.
作者
靳庆文
李胡蓉
徐红霞
Jin Qingwen;Li Hurong;Xu Hongxia(School of Information Resources Management,Renmin University of China,Beijing 100872,China)
出处
《现代情报》
北大核心
2024年第9期59-70,共12页
Journal of Modern Information
关键词
网络欺凌
文本检测
LIME算法
算法解释
数据故事化
cyberbullying
text detection
LIME algorithm
algorithm interpretation
data storytelling
