Polish news articles and their corresponding summaries organized into positive and negative pairs. It utilizes summaries of the same article as positive matches and samples the most semantically similar summaries from different articles to serve as hard negatives for contrastive learning.
Use Cases
- Train contrastive learning models for Polish text similarity using the positive and negative summary pairings
- Develop abstractive summarization systems using the extract_text field as the source input
- Benchmark semantic retrieval algorithms on their ability to distinguish between summaries of the same article versus similar-looking negatives
Strengths
- Focuses exclusively on the Polish language for news summarization and text similarity tasks
- Includes a contrastive structure with positive pairs derived from summaries of the same source article
- Features an 'extract_text' column containing the source article content used for generating summaries