SCARLET: Explainable Attention based Graph Neural Network for Fake News spreader prediction

TL;DR

This paper introduces SCARLET, an explainable graph neural network model that predicts whether nodes in social network information spread are likely to spread false or refutation information, achieving over 87% accuracy.

Contribution

The paper presents a novel GNN model with attention mechanisms that incorporates trust and credibility features for fake news spreader prediction, with explainability of neighborhood feature variations.

Findings

Achieves over 87% accuracy in predicting false information spreaders.

Effectively aggregates trust and credibility features from neighborhood data.

Provides interpretability of feature influence on node behavior.

Abstract

False information and true information fact checking it, often co-exist in social networks, each competing to influence people in their spread paths. An efficient strategy here to contain false information is to proactively identify if nodes in the spread path are likely to endorse false information (i.e. further spread it) or refutation information (thereby help contain false information spreading). In this paper, we propose SCARLET (truSt and Credibility bAsed gRaph neuraL nEtwork model using aTtention) to predict likely action of nodes in the spread path. We aggregate trust and credibility features from a node's neighborhood using historical behavioral data and network structure and explain how features of a spreader's neighborhood vary. Using real world Twitter datasets, we show that the model is able to predict false information spreaders with an accuracy of over 87%.