基于深度学习的ESG“漂绿”风险识别方法  

Deep learning-based ESG“greenwashing”risk identification method

作　　者：叶泞珲 YE Ninghui(School of Economics and Management,Jiangsu University of Science and Technology,Zhenjiang 212000,China)

机构地区：[1]江苏科技大学经济管理学院,江苏镇江212000

出　　处：《江苏科技信息》2025年第6期133-136,共4页Jiangsu Science and Technology Information

摘　　要：在全球金融市场中,Environmental,Social and Governance(ESG)“漂绿”现象频发,误导投资者对企业ESG表现的理解,掩盖真实风险。文章针对传统ESG评估的不足,提出一种基于深度学习的ESG“漂绿”风险识别方法,不依赖评级机构评分,直接处理原始报告。采用“同行相对漂绿得分”作为训练标签,结合TextRank和Ernie-Multi-Head Attention模型,实现了精准的风险识别。在Bloomberg和Wind ESG数据测试集上,该方法平均绝对误差为0.7014,性能优于基线模型。研究有助于投资者深入理解企业ESG绩效,推动可持续发展。In global financial markets,the phenomenon of Environmental,Social,and Governance(ESG)“greenwashing”occurs frequently,misleading investors’understanding of corporate ESG performance and obscuring real risks.To address the limitations of traditional ESG assessments,this paper proposes a deep learning-based ESG“greenwashing”risk identification method that does not rely on rating agency scores but directly processes raw reports.Using“peer-relative greenwashing scores”as training labels and combining TextRank with the Ernie-Multi-Head Attention model,the method achieves accurate risk identification.On the Bloomberg and Wind ESG datasets,the proposed method achieves a mean absolute error of 0.7014,outperforming baseline models.This research helps investors gain a deeper understanding of corporate ESG performance and promotes sustainable development.

关 键 词：ESG评级 漂绿 深度学习 金融投资 Ernie-Multi-Head Attention 

分 类 号：E790.37[军事—军事理论]