Figure 8From: Activism via attention: interpretable spatiotemporal learning to forecast protest activitiesExploration of local and global contributions to forecasting. While the orange nodes represent the states which are correctly predicted by the corresponding models, the gray nodes denote the states either not correctly predicted or where no events occurred, yet still contribute to forecasting events in the correctly predicted states with a value above a certain threshold. The edges indicate the contribution to forecasting from source state to target state. The thicker the edge, the more the contributionBack to article page