Promote to protect: data-driven computational model of peer influence for vaccine perception.

Vaccine hesitancy and acceptance, driven by social influence, is usually explored by most researchers using exhaustive survey-based studies, which investigate public preferences, fundamental values, beliefs, barriers, and drivers through closed or open-ended questionnaires. Commonly used simple statistical tools do not do justice to the richness of this data. Considering the gradual development of vaccine acceptance in a society driven by multiple local/global factors as a compartmental contagion process, we propose a novel methodology where drivers and barriers of these dynamics are detected from survey participants' responses, instead of heuristic arguments. Applying rigorous natural language processing analysis to the survey responses of participants from India, who are from various socio-demographics, education, and perceptions, we identify and categorize the most important factors as well as interactions among people of different perspectives on COVID-19 vaccines. With a goal to achieve improvement in vaccine perception, we also analyze the resultant behavioral transitions through platforms of unsupervised machine learning and natural language processing to derive a compartmental contagion model from the data. Analysis of the model shows that positive peer influence plays a very important role and causes a bifurcation in the system that reflects threshold-sensitive dynamics.