Improve dataset card description and add collection link (#2)

- Improve dataset card description and add collection link (e593177ef37020b5af3c67ece06fcc25a37659fe)


Co-authored-by: Niels Rogge <nielsr@users.noreply.huggingface.co>

README.md

@@ -1,16 +1,22 @@
 ---
-task_categories:
-- text-ranking
-- text-retrieval
 language:
 - en
 size_categories:
 - 1K<n<10K
+task_categories:
+- text-ranking
+- text-retrieval
+license: cc-by-nc-4.0
 ---
 
-Mined hard-negatives using 2 cross encoders with top-k=50
+This dataset contains hard negative examples generated using Large Language Models (LLMs) for training dense retrieval models.  Instead of traditional methods like BM25 or cross-encoders, which require full corpus access, this dataset utilizes an LLM to generate both a query and a hard negative example from a given passage. This corpus-free approach offers a potentially simpler and more efficient alternative for training high-performance retrievers. The dataset includes queries and hard negatives generated using three different methods: an all-LLM pipeline, an LLM-Query->BM25 pipeline, and an LLM-Query->cross-encoder pipeline.  This allows for comparison against traditional methods.
+
+The data was used in the paper [Don't Retrieve, Generate: Prompting LLMs for Synthetic Training Data in Dense Retrieval](https://huggingface.co/papers/2504.21015).
+
+This dataset is part of the Hugging Face collection: [arxiv-hard-negatives-68027bbc601ff6cc8eb1f449](https://huggingface.co/collections/chungimungi/arxiv-hard-negatives-68027bbc601ff6cc8eb1f449)
 
 Please cite if you found this dataset useful 🤗
+
 ```
 @misc{sinha2025dontretrievegenerateprompting,
       title={Don't Retrieve, Generate: Prompting LLMs for Synthetic Training Data in Dense Retrieval},