An Augmented-Anthropology Platform for Real-time Detection, Classification and Mitigation of Infodemics, Fostering Arboviral Vaccine Confidence in the French DROM

"INFODEM-EARLY addresses a paradox at the core of France's epidemic preparedness: the overseas departments and regions (DROM) are both the areas most exposed to emerging arboviruses and the places where national vaccine confidence bottoms out. When La Réunion suspended the live-attenuated chikungunya vaccine (IXCHIQ®) for older adults in April 2025, long-standing anxieties resurfaced and rumours of "créolicide" and "gramounicide" (the elimination of the elderly) spread, showing how rapidly distrust can jeopardise outbreak control. Our project therefore poses a double question: how do arbovirus-related vaccine narratives originate and circulate between online and offline spheres in the DROM, and can a dialogue-centred intervention-delivered by clinicians trained in empathy-restore confidence swiftly enough to safeguard public-health campaigns? To answer it, we combine "augmented anthropology" with state-of-the-art conversational training, the Empathetic Refutational Interview (ERI). First comes augmented anthropology, blending field ethnography with online social listening across mass media (radio, newspapers) and social platforms (X, Facebook, Instagram, YouTube). We will adapt a fine-grained taxonomy of vaccine discourse (> 20 categories) and high-performance large-language-model classifiers (F1 ≈ 0.78) developed for mainland France (ACME, PEPR-MIE) to DROM contexts. Every message will be slotted into one of these categories; any surge of distrust will be characterised and relayed to regional health agencies within twenty-four hours. Ethnographic work-eighty interviews and twenty focus groups in French, Réunion Creole and Shimaore-will anchor these digital traces in lived experience and supply emergent motifs to refine the classifier. In parallel, ERI-recently validated in Europe-will be tailored to island realities. Two four-day masterclasses will train twenty-four clinician-trainers who will disseminate the method to at least 150 frontline professionals in La Réunion (Year 1) and Mayotte (from Month 18). We will also develop ERI-Sim, a secure, browser-based LLM voice agent that simulates hesitant patients so that clinicians can hone their skills before real-world consultations. A mixed-methods evaluation will monitor shifts in clinician confidence, uptake of ERI techniques, ERI-Sim usability, and consultation-room barriers. The project will yield three major advances: the first validated Creole/French LLM classifier for vaccine discourse; the first demonstration of ERI effectiveness beyond continental Europe; and proof that LLM-powered voice agents can scale high-fidelity communication training. All code and training materials will be released under open licences, and local IT teams will be trained during the final year to ensure sustainable regional ownership. By coupling real-time listening with empathic dialogue, INFODEM-EARLY offers a transferable model for countering infodemics during future arbovirus vaccination campaigns in the DROM and other affected regions of the Global South."