50 clusters of news articles from Wikinews containing 2,088 sentences annotated with 18 Cross-document Structure Theory (CST) relationship types. The data identifies semantic links such as 'Identity', 'Subsumption', and 'Overlap' across multiple documents covering the same news events.
Use Cases
- Train a classifier to identify 'Contradiction' or 'Overlap' between sentence pairs using the relationship labels.
- Develop multi-document summarization algorithms that utilize 'Subsumption' annotations to filter redundant content.
- Evaluate cross-document coreference resolution systems using the 'Identity' relationship markers.
Strengths
- 18 distinct semantic relationship labels including 'Identity', 'Subsumption', and 'Contradiction'.
- 50 thematic clusters of articles sourced from the Wikinews platform.
- 2,088 sentences manually aligned to map semantic overlaps and differences between independent news reports.