A Meta Path-based Approach for Rumor Detection on Social Media

TL;DR

This paper introduces MGLAN, a novel meta path-based neural network that effectively detects rumors on social media by leveraging the heterogeneous structure of rumor propagation networks, outperforming existing models.

Contribution

The paper proposes a new meta path-based embedding approach integrated into a neural network for rumor detection, exploiting the heterogeneous nature of social media propagation networks.

Findings

MGLAN outperforms existing models on three datasets.

Meta path-based embeddings improve node discrimination.

Structural features enhance rumor detection accuracy.

Abstract

The prominent role of social media in people's daily lives has made them more inclined to receive news through social networks than traditional sources. This shift in public behavior has opened doors for some to diffuse fake news on social media; and subsequently cause negative economic, political, and social consequences as well as distrust among the public. There are many proposed methods to solve the rumor detection problem, most of which do not take full advantage of the heterogeneous nature of news propagation networks. With this intention, we considered a previously proposed architecture as our baseline and performed the idea of structural feature extraction from the heterogeneous rumor propagation over its architecture using the concept of meta path-based embeddings. We named our model Meta Path-based Global Local Attention Network (MGLAN). Extensive experimental analysis on three state-of-the-art datasets has demonstrated that MGLAN outperforms other models by capturing node-level discrimination to different node types.