Rumour detection

Rumour detection is the task of identifying unverified claims—especially those that are apparently credible but hard to verify and produce skepticism or anxiety in communities—that circulate on social media or online platforms.

Definition

A rumour is a circulating story of questionable veracity that is: - Apparently credible but difficult to verify - Produces sufficient skepticism and/or anxiety - Motivates people to seek the truth

Task components

Rumour detection and analysis typically involves:

1. Identification: Detecting which claims in social media are rumours
2. Stance classification: Understanding how users respond to the rumour
3. Veracity assessment: Determining if the rumour is true, false, or unverifiable
4. Tracking: Following how the rumour spreads and evolves over time

Applications

• Journalism: Real-time fact-checking during breaking news
• Crisis response: Identifying false information during emergencies (natural disasters, public health crises)
• Social media moderation: Flagging unverified claims for review
• Decision support: Helping platforms and individuals assess information reliability

Challenges

• Class imbalance: True claims, false claims, and unverifiable claims have different distributions
• Context dependency: Veracity often depends on external knowledge, temporal information, and domain expertise
• Community dynamics: Response patterns (support, denial, questioning) vary by rumour and event
• Early detection: Verifying rumours in real-time, before resolution evidence emerges

Related work

Key papers and benchmarks

• Zubiaga et al. (2016) — Stance Classification in Rumours as a Sequential Task Exploiting the Tree Structure of Social Media Conversations — introduces Tree CRF to exploit conversational structure for stance; demonstrates that sequential models outperform non-sequential baselines
• Kochkina et al. (2017) — Sequential Approach to Rumour Stance Classification with Branch-LSTM — state-of-the-art neural approach for conversation-aware stance classification on RumourEval 2017
• Detection and Resolution of Rumours in Social Media: A Survey — comprehensive survey of rumour detection pipeline components (detection, tracking, stance, veracity), datasets, annotation schemes, and approaches
• Kochkina et al. (2018) — All-in-one: Multi-task Learning for Rumour Verification — demonstrates multi-task learning improvements for veracity classification by jointly training with stance and detection tasks
• RumourEval 2019 — extended shared task with Twitter and Reddit data; 22 systems; demonstrates progress via contextual embeddings and ensemble approaches
• SemEval-2017 Task 8: RumourEval — benchmark task with datasets and shared challenge results
• Pheme Project — foundational research on rumour analysis in social networks