Enhancing Language Models for Robust Greenwashing Detection
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912859211956224 |
|---|---|
| author | Braun, Neil Heinrich Ong, Keane Mao, Rui Cambria, Erik Mengaldo, Gianmarco |
| author_facet | Braun, Neil Heinrich Ong, Keane Mao, Rui Cambria, Erik Mengaldo, Gianmarco |
| contents | Sustainability reports are critical for ESG assessment, yet greenwashing and vague claims often undermine their reliability. Existing NLP models lack robustness to these practices, typically relying on surface-level patterns that generalize poorly. We propose a parameter-efficient framework that structures LLM latent spaces by combining contrastive learning with an ordinal ranking objective to capture graded distinctions between concrete actions and ambiguous claims. Our approach incorporates gated feature modulation to filter disclosure noise and utilizes MetaGradNorm to stabilize multi-objective optimization. Experiments in cross-category settings demonstrate superior robustness over standard baselines while revealing a trade-off between representational rigidity and generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_21722 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Enhancing Language Models for Robust Greenwashing Detection Braun, Neil Heinrich Ong, Keane Mao, Rui Cambria, Erik Mengaldo, Gianmarco Computation and Language Artificial Intelligence Sustainability reports are critical for ESG assessment, yet greenwashing and vague claims often undermine their reliability. Existing NLP models lack robustness to these practices, typically relying on surface-level patterns that generalize poorly. We propose a parameter-efficient framework that structures LLM latent spaces by combining contrastive learning with an ordinal ranking objective to capture graded distinctions between concrete actions and ambiguous claims. Our approach incorporates gated feature modulation to filter disclosure noise and utilizes MetaGradNorm to stabilize multi-objective optimization. Experiments in cross-category settings demonstrate superior robustness over standard baselines while revealing a trade-off between representational rigidity and generalization. |
| title | Enhancing Language Models for Robust Greenwashing Detection |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2601.21722 |