Abstract

“Fake news” refers to news articles that are made-up stories with an intention to deceive. This is not a new problem as it has occurred since the very first years of the printing press.
However, it attracted a growing interest with the wide use of social media streams as online news sources.
Despite the recent efforts, the fake news detection problem is far from being solved since the performance in tasks such as FEVER [1] is quite limited (74% FEVER score), and the performance drops even further in real life scenarios (49% F1, Multi-FC [2]).
Our recent work addresses different aspects of the problem: we work on fact extraction and verification [3] which includes natural language processing methods that perform automated fact checking. Furthermore, we develop graph-neural network methods that explore the correlation between diverse news items in predicting their veracity [4].
We also develop deep learning models that explore effectively the propagation pattern of news in social media, including the sequence of interactions of social media users with news across time [5]

[1] Thorne, James, et al. "FEVER: a Large-scale Dataset for Fact Extraction and VERification." NAACL 2018
[2] Augenstein, Isabelle, et al. "MultiFC: A Real-World Multi-Domain Dataset for Evidence-Based Fact Checking of Claims." EMNLP 2019
[3] G. Bekoulis, C. Papagiannopoulou, N. Deligiannis, Understanding the Impact of Evidence-AwareSentence Selection for Fact Checking, NLP4IF@NAACL 2021
[4] Nguyen, Duc Minh, et al. "Fake news detection using deep markov random fields." NAACL 2019
[5] T. D. Huu, X. Luo, D. M. Nguyen, N. Deligiannis, “Rumour detection via news propagation dynamics and user representation learning” IEEE Data Science Workshop, DSW’19, 2019.