Quantifying Temporal Novelty in Social Networks Using Time-Varying Graphs and Concept Drift Detection

Abstract

This paper presents a new approach to quantify temporal novelties in Social Networks and, as a consequence, to identify changing points driven by the occurrence of new real-world events that influence the public opinion. Our approach starts using Text Mining tools to highlight the main key terms, that will be later used to create a temporal graph, thus preserving their relation into the original texts and their temporal dependencies. We also defined a new measure to quantify the way users’ opinions have been evolving over time. Finally, we propose a straightforward Concept Drift method to identify when the changing points happen. Our full approach was evaluated on a historical event in Brazil: the 2018 presidential election race. We have chosen this period due to the volume of publications that, definitely, stated Social Networks as the main mechanism for new political activism. Our good results emphasize the importance of our approach and open new possibilities to identify bots developed to just spread, for example, fake news.