Vaccine Confidence and Infodemic in Southeast Asia’s Nusantara Sociocultural Sphere: Human-AI Collaboration in Infodemiology and Prebunking as Solutions

Background: The COVID-19 pandemic represented a dual public health crisis: a viral outbreak and an infodemic that undermined vaccine confidence through conflicting and misleading information. In Southeast Asia’s Nusantara sociocultural sphere, which includes Brunei, Indonesia, Malaysia, Singapore, and adjacent communities, shared use of Bahasa Melayu and Bahasa Indonesia among 290 million speakers facilitates the rapid spread of digital content, including misinformation. Yet vaccine discourse in these languages remains underrepresented in computational research and prebunking intervention design.Objective: This three-part study examined digital trends in vaccine confidence and developed a culturally resonant prebunking intervention for Bahasa Melayu and Bahasa Indonesia speakers across the Nusantara region. We aimed to (1) evaluate the capacity of a large language model (LLM) to analyze sentiment and themes relevant to vaccine confidence, (2) assess trust and mistrust dynamics across platforms (2020–2023), and (3) co-create and refine a technique-based prebunking video using a human-centered design (HCD) approach.Methods: Paper 1 benchmarked eight prompting strategies using OpenAI’s GPT-4o on 248 annotated captions. Paper 2 applied the top-performing prompt (few-shot + chain-of-thought or FS + CT) to 11,119 captions from Facebook and Telegram to analyze temporal shifts in vaccine discourse. Paper 3 used HCD workshops with participants from Brunei, Malaysia, and Indonesia, followed by focus group discussions (FGDs) in Banda Aceh (Indonesia), Kota Kinabalu (East Malaysia), and Kuala Terengganu (West Malaysia) to iteratively refine a video prototype.Results: The FS+CT prompt achieved the highest F1 scores for sentiment (0.916) and thematic classification (0.746). Applying this validated strategy to the full dataset revealed a rise in neutral sentiment by 2023 and recurring overlaps between trust and mistrust expressions. Among the most viral content, 80.0% of Facebook captions expressed mistrust in both vaccine ...