Update README.md

README.md

@@ -1,4 +1,16 @@
 ---
+language:
+- en
+multilinguality:
+- monolingual
+size_categories:
+- 100K<n<1M
+task_categories:
+- feature-extraction
+- sentence-similarity
+pretty_name: Natural Questions with mined hard negatives
+tags:
+- sentence-transformers
 dataset_info:
 - config_name: triplet-5
   features:
@@ -46,3 +58,41 @@ configs:
   - split: train
     path: triplet-all/train-*
 ---
+
+# Dataset Card for Natural Questions with mined hard negatives
+
+This dataset is a collection of question-answer-negative triplets and question-answer-negative_1...-negative_5 tuples from the Natural Questions dataset. See [Natural Questions](https://huggingface.co/datasets/sentence-transformers/natural-questions) for additional information.
+This dataset can be used directly with Sentence Transformers to train embedding models.
+
+The negative samples have been automatically mined with [all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) and:
+* `range_min`: 10, i.e. we skip the 10 most similar samples
+* `range_max`: 20, i.e. we only look at the top 20 most similar samples
+* `margin`: 0, i.e. we require negative similarity + margin < positive similarity, so negative samples can't be more similar than the known true answer
+* `sampling_strategy`: "random", i.e. we randomly sample from the candidate negatives rather than taking the "top" negatives
+* `num_negatives`: 5, i.e. we mine 5 negatives per question-answer pair
+
+## Dataset Subsets
+
+### `triplet` subset
+
+* Columns: "question", "answer", "negative"
+* Column types: `str`, `str`, `str`
+* Examples:
+    ```python
+    {
+    }
+    ```
+* Collection strategy: Reading the natural questions dataset from [sentence-transformers/natural-questions](https://huggingface.co/datasets/sentence-transformers/natural-questions) and performing hard negative mining with `as_triplets=True`.
+* Deduplified: No
+
+### `triplet-5` subset
+
+* Columns: "question", "answer", "negative_1", "negative_2", "negative_3", "negative_4", "negative_5"
+* Column types: `str`, `str`, `str`, `str`, `str`, `str`, `str`
+* Examples:
+    ```python
+    {
+    }
+    ```
+* Collection strategy: Reading the natural questions dataset from [sentence-transformers/natural-questions](https://huggingface.co/datasets/sentence-transformers/natural-questions) and performing hard negative mining with `as_triplets=False`.
+* Deduplified: No