A Graph Temporal Information Learning Framework for Popularity Prediction

Effectively predicting the future popularity of online content has important implications in a wide range of areas, including online advertising, user recommendation, and fake news detection. Existing approaches mainly consider the popularity prediction task via path modeling or discrete graph modeling. However, most of them heavily exploit underlying diffusion structural and sequential information, while ignoring the temporal evolution information among different snapshots of cascades. In this paper, we propose a graph temporal information learning framework based on an improved graph convolutional network (GTGCN), which can capture both the temporal information governing the spread of information in a snapshot, and the inherent temporal dependencies among different snapshots. We validate the effectiveness of the GTGCN by applying it on a Sina Weibo dataset in the scenario of predicting retweet cascades. Experimental results demonstrate the superiority of our proposed method over the state-of-the-art approaches.