Mean Birds: Twitter Aggression and Bullying Detection with 1.6M Tweets

Despoina Chatzakou from Aristotle University of Thessaloniki presents a dataset for detecting bullying and aggressive behavior on Twitter. The corpus contains 1.6 million tweets posted over a 3-month period. The research proposes a methodology extracting text, user, and network-based attributes to distinguish bullies and aggressors from regular users.

Use Cases

• Train classifiers to detect bullying behavior based on user posting patterns and network features.
• Analyze the characteristics of aggressors based on the negativity and popularity metrics described.
• Study the relationship between online community participation and bullying behavior as suggested by the findings.

Strengths

• Corpus of 1.6 million tweets provides a substantial text base for analysis.
• Methodology evaluated with machine learning models achieving over 90% AUC, indicating strong predictive performance.
• Attributes derived from text, user, and network data offer a multi-faceted feature set.

Limitations

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

Provenance

Source

Aristotle University of Thessaloniki

Collection Method

Data likely gathered from Twitter's public API, with attributes extracted for user classification.

Time Range

Tweets posted over a 3-month period (specific dates unknown).