Event Prediction in Complex Social Graphs via Feature Learning of Vertex Embeddings

Abstract

Social network graphs and structures possess implicit and latent knowledge about their respective vertices/actors and edges/links which may be exploited, using effective and efficient techniques, for predicting events within the social graphs. Understanding the intrinsic relationship patterns among spatial social actors as well as their respective properties are very crucial factors to be taken into consideration with regard to event prediction in social network graphs. Thus, in this paper, we model an event prediction task as a classification problem; and we propose a unique edge sampling approach for predicting events in social graphs by learning the context of each actor via neighboring actors/nodes with the goal of generating vector-space embeddings per actor. Successively, these relatively low-dimensional node embeddings are fed as input features to a downstream classifier for event prediction about the reference social graph. Training and evaluation of our approach was done on popular and real-world datasets, viz: Cora and Citeseer.