FANG

We propose Factual News Graph (FANG), a novel graphical social context representation and learning framework for fake news detection. Unlike previous contextual models that have targeted performance, our focus is on learning. Compared to transductive models, FANG scalable in training as it does not maintain the entities involved propagation of other efficient at inference time, without need reprocess entire graph. Our experimental results show better capturing into high-fidelity representation, compared recent nongraphical models. In particular, yields significant improvements task detection robust case limited data. further demonstrate representations learned by generalize related tasks, such predicting factuality reporting medium.