Listening at scale: how social media informs behavioural trends in real time

Abstract Social media analysis offers valuable insights into public health-related behaviours, such as attitudes toward vaccination and adherence to non-pharmaceutical interventions (e.g., mask-wearing), as well as determinants like trust in institutions and perceived risk. These digital traces help identify disinformation trends, public doubts, information needs, and behavioural signals - elements that could be integrated into behavioural models. The University of Pisa developed a structured social media monitoring approach through the RISP project (2023-2024), based on the WHO and UNICEF Infodemic Insights Reporting methodology. This system combines automated dashboards with manual narrative analysis, applied monthly to vaccine-related content. As an example, we will present insights from flu vaccine monitoring during the 2023-2024 season, which showed a rise in discouraging stances as institutional communication declined, alongside recurring questions on eligibility and safety. This monitoring approach forms the basis for ongoing work in the BEHAVE_MOD project, where social media data will be used to explore and model health-related behaviours across diseases such as flu, RSV, and Dengue. To make social media monitoring more directly usable in modelling frameworks, we are working to go beyond qualitative narrative identification by developing automated tools for stance classification. Specifically, we are building a framework that uses natural language processing (NLP) and large language models (LLMs) to classify vaccine stance at scale. This enables the generation of structured behavioural indicators-such as stance proportions over time-that can be quantified and integrated into mathematical models. While social media analysis cannot replace traditional behavioural surveillance systems, it can complement them by providing a low-cost, scalable, and near real-time data stream to support public health monitoring, communication, and modelling in routine and emergency settings.