Abstract

While prior crowdfunding research emphasizes interpersonal dynamics, this study examines the Linguistic Hierarchy of Trust in ESG-oriented digital finance—a critical driver of sustainable e-commerce. Integrating Signaling Theory (ST) and Language Expectancy Theory (LET), the research examines the structural linguistic mechanisms fundraisers use to establish legitimacy and mitigate “greenwashing” risks in AI-intensive environments. Analysis of 1,420 campaigns from FlyingV and Zeczec employs a multi-stage Natural Language Processing (NLP) experiment. Results identify a “Structural Dilution Effect” in traditional high-dimensional models (e.g., TF-IDF), where theoretically potent ESG signals are systematically overwhelmed by lexical noise. Transitioning to a theory-driven, parsimonious feature set yields substantial and stable accuracy gains: a Random Forest (RF) classifier proves most reliable, attaining the highest mean accuracy (0.83) with the lowest variance across random states, while a Support Vector Machine (SVM) reaches the single highest peak of 91%. These results demonstrate that “less is more” when feature engineering is grounded in established communication theory. The study advances a theoretical account—the Linguistic Hierarchy of Trust—in which concrete, verifiable language provides the foundation on which higher-order ESG claims must rest to avoid being discounted as “cheap talk.” Empirically, the analysis isolates two mechanisms: a structural dilution effect, whereby ESG signals lose predictive value when concatenated with high-dimensional TF-IDF vectors, and a parsimony effect, whereby a compact set of ESG signal intensity and sentiment features restores that predictive value. This study provides an NLP-driven governance framework for e-commerce platforms to enhance signal integrity, reduce epistemic uncertainty, and promote a sustainable digital ecosystem in the era of generative AI.

Download

Share

COinS