Paper Key : IRJ************197
Author: Sourabh Chaudhari,Anjali S. Khandagale,Rajnandini Jadhav,Harsh Gulhane,Md. Owais Khan
Date Published: 01 Mar 2023
Abstract
Social media has become a major source of news for people all over the world due to its ease of access, cost-effectiveness, and swift distribution. However, this also leads to concerns about the credibility of the information and exposure to false news, which is created with the intention of deceiving readers. Detecting false news automatically is a challenging task that goes beyond the capability of traditional content-based analytical tools. This is because the interpretation of news often requires an understanding of political or social context and common sense, which current natural language processing algorithms lack. Recent studies have shown that false and authentic news have different patterns of propagation on social media, which can be used to detect false news automatically. Propagation-based techniques have advantages over content-based methods, such as being language-independent and more resilient to adversarial attacks. In this research, we present a unique automatic false news detection algorithm based on geometric deep learning. This algorithm is a graph-based extension of standard convolutional neural networks, and can integrate heterogeneous data such as content, user profile and behavior, social graph, and news dissemination. The algorithm was trained and tested on news articles shared on Twitter that have been validated by professional fact-checking groups, with results showing that social network structure and propagation are crucial factors in accurately detecting false news (92.7% ROC AUC). Additionally, we found that false news can be successfully identified early on, even after only a few hours of spread. Finally, we evaluated the aging of our model using training and testing data separated by time and found that propagation-based techniques have potential as an alternative or complementary strategy for false news identification.