[Paper Review] Beyond News Contents: The Role of Social Context for Fake News Detection

Kai Shu, Suhang Wang+1 more|arXiv (Cornell University)|Dec 20, 2017

Misinformation and Its Impacts34 references106 citations

TL;DR

TriFN leverages publisher-news relationships and user engagements to detect fake news, outperforming baselines on FakeNewsNet (BuzzFeed and PolitiFact).

ABSTRACT

Social media is becoming popular for news consumption due to its fast dissemination, easy access, and low cost. However, it also enables the wide propagation of fake news, i.e., news with intentionally false information. Detecting fake news is an important task, which not only ensures users to receive authentic information but also help maintain a trustworthy news ecosystem. The majority of existing detection algorithms focus on finding clues from news contents, which are generally not effective because fake news is often intentionally written to mislead users by mimicking true news. Therefore, we need to explore auxiliary information to improve detection. The social context during news dissemination process on social media forms the inherent tri-relationship, the relationship among publishers, news pieces, and users, which has potential to improve fake news detection. For example, partisan-biased publishers are more likely to publish fake news, and low-credible users are more likely to share fake news. In this paper, we study the novel problem of exploiting social context for fake news detection. We propose a tri-relationship embedding framework TriFN, which models publisher-news relations and user-news interactions simultaneously for fake news classification. We conduct experiments on two real-world datasets, which demonstrate that the proposed approach significantly outperforms other baseline methods for fake news detection.

Motivation & Objective

• Motivate fake news detection in the social media era where content alone is insufficient.
• Define a tri-relationship (publishers, news, users) to model propagation dynamics.
• Develop TriFN to learn joint representations from contents, social context, and interactions.
• Demonstrate effectiveness through real-world dataset experiments.

Proposed method

• Proposes TriFN, a tri-relationship embedding framework combining five components: news contents embedding, user embedding, user-news interaction embedding, publisher-news relation embedding, and semi-supervised classification.
• Uses nonnegative matrix factorization (NMF) to embed news contents and users (D,V) and (U,T) respectively.
• Incorporates user credibility scores to guide user-news interaction learning via weighted distance terms linking user latent features to true/fake labels.
• Models publisher partisan bias with labeled publishers via a publisher-news relation embedding that aligns publisher features with partisan scores.
• Implements a semi-supervised classifier with a linear mapping from news latent features to labels, optimized via alternating least squares across D,U,V,T,p,q.

Experimental results

Research questions

• RQ1 Can TriFN improve fake news classification by jointly modeling publisher partisan bias and user engagements?
• RQ2 How do publisher bias modeling and user engagement learning contribute to TriFN's performance?
• RQ3 Can TriFN enable early fake news detection when user engagements are limited?

Key findings

• TriFN significantly outperforms baselines on two real-world datasets (BuzzFeed and PolitiFact) in F1 scores and other metrics.
• On BuzzFeed, TriFN achieves an F1 of 0.870, outperforming RST (0.633) and LIWC+Castillo (0.822).
• On PolitiFact, TriFN achieves an F1 of 0.880, outperforming RST (0.615) and LIWC+Castillo (0.843).
• The framework leverages publisher partisan bias and user credibility to improve news representations and fake news prediction.
• Experiments are conducted on FakeNewsNet with 80-20 train-test splits, averaged over 10 runs.