Hybrid AI Framework for Environmental Misinformation Detection

The proliferation of environmental misinformation on social media poses significant challenges for public awareness and policy-making. While large language models (LLMs) and knowledge-based reasoning have advanced fake-news detection, most approaches remain domain-agnostic and lack semantic interpretability. This paper proposes a hybrid framework that integrates generative AI with domain-specific ontologies to detect and explain environmental misinformation. The framework comprises: (i) claim extraction with GPT-4, (ii) semantic alignment using environmental ontologies (ENVO, GEMET, AGROVOC), (iii) hybrid classification that fuses linguistic and symbolic features, and (iv) evaluation through quantitative metrics and ontology-driven consistency checks. Experiments on a curated dataset of 2,200 English-language social media posts show that our system outperforms transformer-only and ontology-only baselines in precision, recall, and F1, while providing transparent, ontology-grounded rationales. We also discuss key limitations—dataset size, monolingual coverage, and LLM reproducibility—as well as scalability concerns for real-world deployment. Future work includes multilingual extensions, ontology enrichment for emerging ecological concepts, and the use of lightweight or open-source LLMs to improve cost-efficiency and reproducibility.