How can media attention reveal ESG improvement opportunities? A multi-algorithm machine learning-based approach for Taiwan’s electronics industry

The wave of discussions on ESG (environment, social, governance) issues widely suggest that ESG goals can benefit companies and provide corresponding advantages to investors. However, few studies consider the actual value that ESG performance can deliver, leading to overly high expectations regarding ESG investments (Cornell & Damodaran, 2020). Companies with high ESG expectations may overinvest in such initiatives. To counteract the potential biases this could introduce, ESG ratings agencies might discreetly adjust their weighting methods to ensure more accurate assessments. Owning to differing focal points among stakeholders, ESG scores lack persuasive reform suggestions for corporations to improve ESG actions, reducing corporate enthusiasm and confidence in ESG resource allocation. This study employs the Refinitiv news database and multi-algorithm machine learning methods to target the ESG scores of Taiwan-listed companies in the electronics industry. Neural networks (NN), support vector machine (SVM) learning, and random forest algorithms are used to construct a multi-algorithm machine learning-based approach to explore the predictive ability of media attention.

The results show that specific ESG practices of listed corporations in Taiwan’s electronics industry need to be strengthened, especially regarding CSR (corporate social responsibility) strategies and Human Rights. Media attention positively impacted the comprehensive ESG scores of companies in Taiwan’s electronics industry; however, the impact on individual companies was inconsistent. Finally, integrating stacked generalization models can improve the prediction accuracy of ESG analysis; the contribution of support vector machine algorithms was most prominent in the study sample.

This study uses multi-algorithm machine learning methods to establish a prediction model for the impact of the comprehensive ESG scores within Taiwan’s listed electronics industry on media attention. The empirical findings suggest the practical application value of the semi-supervised learning and integrated stacking generalization models for conducting ESG research; exploring enterprise shortcomings in ESG investment allocation provides feedback for enterprise planning via ESG management decision-makers and resource allocation. We recommend that listed corporations in Taiwan’s electronics industry prioritize issues related to Human Rights and Emissions Reduction when implementing future ESG actions. This study suggests potential directions for future ESG research, such as employing semi-supervised learning or support vector machines to enhance research methods.