PHEME: Twitter Rumours and Non-Rumours from Five Breaking News Events

PHEME dataset contains a collection of Twitter rumours and non-rumours posted during five breaking news events. The dataset includes 1,972 rumours and 3,830 non-rumours across events like the Charlie Hebdo attack, Ferguson unrest, and Germanwings Crash. It was created by Arkaitz Zubiaga for the paper 'Learning Reporting Dynamics during Breaking News for Rumour Detection in Social Media'.

Use Cases

• Train classifiers for rumour detection based on tweet content and conversational threads.
• Analyze the spread and dynamics of information versus misinformation during breaking news events.
• Study the structure of social media conversations in rumours versus non-rumours based on the provided source tweet and reaction tweet directories.

Strengths

• Provides a clear, event-based structure with separate directories for rumours and non-rumours.
• Contains 5,802 total annotated tweets across five distinct breaking news events.
• Includes conversational context with source tweets and their reaction threads.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset including reaction tweets is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Arkaitz Zubiaga

Collection Method

Collected from Twitter during breaking news events and annotated as rumours or non-rumours.