SBIR/STTR Award attributes
Thresher proposes reverse engineering the governments' crisis communications playbooks so that we can predict which online communication strategy a given government will use in a given crisis and inform a USG counter playbook. Our analysts have identified several communications “plays” (i.e. online interventions designed to manipulate the online conversation) in our dataset. Our data set includes responses to local and global crises. The case study approach is the most common approach to understanding crisis communication in social media, and those longitudinal studies that have been conducted only examine about 100 cases. Based on our analysis of online communication patterns during several crises, we hypothesize that governments are using nuanced plays that can be detected automatically by analyzing our labeled data at scale. We will combine our experience applying neural network language models to isolate narratives in events with techniques in signal processing and multivariate time series clustering in order to automatically catalog plays. We will then develop models to predict which play will be used in a given crisis using supervised machine learning and change point detection. We will test if and how the playbook applies across social media platforms.
