We study how emojis are used to express solidarity in social media in the context of two major crisis events - a natural disaster, Hurricane Irma in 2017 and terrorist attacks that occurred on November 2015 in Paris. Using annotated corpora, we first train a recurrent neural network model to classify expressions of solidarity in text. Next, we use these expressions of solidarity to characterize human behavior in online social networks, through the temporal and geospatial diffusion of emojis. Our analysis reveals that emojis are a powerful indicator of sociolinguistic behaviors (solidarity) that are exhibited on social media as the crisis events unfold.