Efficient Semantic Detection and Analysis of Misinformation in CBD-Related Tweets Using FAISS and Mistral NeMo Instruct

The growing popularity of cannabidiol (CBD) has led to a surge in misinformation, particularly on social media platforms like Twitter, posing risks to public health. This paper presents a scalable method for detecting CBD-related misinformation in a large corpus of tweets. Using approximately 3.7 million tweets collected from 2011 to 2021, we implement a two-step process: first, FAISS (Facebook AI Similarity Search) efficiently identifies tweets semantically similar to false claims extracted from FDA warning letters. Second, Mistral NeMo Instruct, a zero-shot model, classifies tweets as ‘Misinformation’ or ‘Non-Misinformation’, providing justifications for transparency. This approach minimizes computational costs while maintaining accuracy, making it a practical tool for large-scale misinformation detection. The framework is scalable and adaptable, evolving with new FDA data or emerging cannabis research.