Figure 6From: Activism via attention: interpretable spatiotemporal learning to forecast protest activitiesMean absolute values of intra-region and inter-region input (gate) weights. These are the input weights learned from the neural network model (the LSTM networks in the temporal and spatial components) and the magnitude of weights (which can take any values) allows for a comparison of the relative importance of different features. (a) Intra-region input weights. (b) Inter-region input weightsBack to article page