Fig. 8From: Task-agnostic representation learning of multimodal twitter data for downstream applicationsLocation prediction task. The input tweet’s component and user location embeddings are concatenated and input to a classifier that predicts the tweet’s locationBack to article page