MNLP_M3_rag_dataset / README.md
emiliensilly's picture
Upload README.md with huggingface_hub
270bd08 verified
# STEM Wikipedia Subset
This dataset was created by filtering the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dump to retain only science-related articles.
## Filtering Process
1. A handpicked list of ~80 STEM keywords was defined.
2. Each Wikipedia article's `text` field was scanned for these keywords.
3. A count of keyword matches was computed for each article.
4. Only articles with ≥ X keyword hits were retained.
5. Only the `text` and `source` columns were preserved.
## Use Case
This dataset is suitable for pretraining or domain adaptation of language models focused on STEM reasoning or QA tasks.