Detecting and Forecasting Misinformation via Temporal and Geometric Propagation Patterns

EasyChair Preprint 5068

Abstract

Misinformation generally takes the form of a false or un-verified claim under the guise of fact. It is necessary to protect socialmedia sites against the spread of misinformation by means of effectivemisinformation detection and analysis. To this end, we first formulatemisinformation propagation as a dynamic graph. Then we extract thetemporal evolution patterns and geometric features of the propagationgraph based on Temporal Point Processes (TPPs). TPPs provide the ap-propriate modelling framework for a list of stochastic, discrete events. Inthis context, that is a sequence of social user engagements. Furthermore,we forecast the cumulative number of engaged users based on a powerlaw. Such forecasting capabilities can be useful in assessing the threatlevel of an individual piece of misinformation. By jointly considering thegeometric and temporal dynamics, our model has achieved comparableperformance with state-of-the-art baselines on two well known datasets

Keyphrases: Propagation Graph, misinformation, point processes

@booklet{EasyChair:5068,
  author    = {Qiang Zhang and Jonathan Cook and Emine Yilmaz},
  title     = {Detecting and Forecasting Misinformation via Temporal and Geometric Propagation Patterns},
  howpublished = {EasyChair Preprint 5068},
  year      = {EasyChair, 2021}}