[Paper Review] FakeNewsNet: A Data Repository with News Content, Social Context and Spatialtemporal Information for Studying Fake News on Social Media

Kai Shu, Deepak Mahudeswaran+3 more|arXiv (Cornell University)|Sep 5, 2018

Misinformation and Its Impacts29 references215 citations

TL;DR

The paper introduces FakeNewsNet, a multi-dimensional data repository that combines news content, social context, and spatiotemporal information to study fake news on social media, and demonstrates datasets, analyses, and baseline detection results.

ABSTRACT

Social media has become a popular means for people to consume news. Meanwhile, it also enables the wide dissemination of fake news, i.e., news with intentionally false information, which brings significant negative effects to the society. Thus, fake news detection is attracting increasing attention. However, fake news detection is a non-trivial task, which requires multi-source information such as news content, social context, and dynamic information. First, fake news is written to fool people, which makes it difficult to detect fake news simply based on news contents. In addition to news contents, we need to explore social contexts such as user engagements and social behaviors. For example, a credible user's comment that "this is a fake news" is a strong signal for detecting fake news. Second, dynamic information such as how fake news and true news propagate and how users' opinions toward news pieces are very important for extracting useful patterns for (early) fake news detection and intervention. Thus, comprehensive datasets which contain news content, social context, and dynamic information could facilitate fake news propagation, detection, and mitigation; while to the best of our knowledge, existing datasets only contains one or two aspects. Therefore, in this paper, to facilitate fake news related researches, we provide a fake news data repository FakeNewsNet, which contains two comprehensive datasets that includes news content, social context, and dynamic information. We present a comprehensive description of datasets collection, demonstrate an exploratory analysis of this data repository from different perspectives, and discuss the benefits of FakeNewsNet for potential applications on fake news study on social media.

Motivation & Objective

• Motivate the need for comprehensive fake news datasets that include content, social context, and spatiotemporal information.
• Describe the construction and public release of FakeNewsNet with two datasets and rich features.
• Demonstrate exploratory analyses to characterize data properties and baseline fake news detection performance.
• Discuss the potential applications and research opportunities enabled by the repository.

Proposed method

• Integrate multi-dimensional data from news content, social context, and spatiotemporal information using an end-to-end pipeline (FakeNewsTracker).
• Curate ground-truth labels for fake and real news from fact-checking sources (PolitiFact, GossipCop) and recover content via archives or web search when needed.
• Assemble extensive social context data (user profiles, posts, and network information) by collecting social engagements and metadata from platforms (e.g., Twitter).
• Extract and summarize content features (linguistic, visual) and contextual signals (social behavior, engagement patterns).
• Perform baseline fake news detection using multiple models on content alone, social context, and their fusion (SAF variants).
• Provide an API and data structure for efficient access and subset retrieval of the large repository.

Experimental results

Research questions

• RQ1 What are the characteristics and signals across content, social context, and spatiotemporal dimensions that differentiate fake from real news?
• RQ2 How does incorporating social context and temporal information improve fake news detection performance compared to content-only models?
• RQ3 What baseline benchmarks and features can guide future research using a multi-dimensional fake news dataset?
• RQ4 How can the repository support early fake news detection and diffusion studies?

Key findings

• FakeNewsNet combines news content, social context, and spatiotemporal data from PolitiFact and GossipCop to enable multi-faceted fake news research.
• Content-based models yield around 65-66% accuracy on PolitiFact and GossipCop baselines, with fusion models (SAF) generally improving performance.
• Social-context features (engagement patterns) plus content improve detection; SAF (Social Article Fusion) achieves up to 0.691 accuracy on PolitiFact and 0.792 F1 on GossipCop in the reported results.
• Bot analysis shows higher presence of bots among fake-news-related users compared to real-news-related users, with notable differences in engagement types (replies vs. retweets).
• Temporal patterns reveal that fake news tends to exhibit rapid retweet spikes and fewer replies relative to real news, suggesting potential for early detection signals.
• The paper provides a scalable API and data format to access subsets of the large dataset, facilitating reproducibility and reuse.