Maven Event Dataset: Metadata and 168 Event Labels for ACE Detection

Metadata for the ACE (Automatic Content Extraction) event detection dataset, featuring 168 event type labels and 5 task stream configurations. Created by datht and updated in February 2026, it provides the structural mapping and frequency distributions necessary for continual learning research.

Use Cases

• Benchmarking continual learning models using the task configurations in streams.json
• Mapping model predictions to event types using label2id.json
• Analyzing label sparsity and distribution via label_freq.json

Strengths

• 168 distinct event type labels for high-granularity classification
• 5 pre-defined task streams for standardized continual learning benchmarking
• Includes frequency distribution data for 168 keys to assist with class imbalance

Limitations

• Metadata only; lacks raw text corpus required for model training
• Small number of task streams (5) limits longitudinal study depth

Provenance

Source

ACE (Automatic Content Extraction) program

Collection Method

annotated

Freshness

Last updated February 2026.