Improve dataset card: document schema, measured stats and limitations

README.md

@@ -28,8 +28,58 @@ size_categories:
 - 1K<n<10K
 ---
 
-# Central Kurdish Sentiment Dataset
-This Central Kurdish dataset is compiled from the [transfer learning project for sentiment analysis](https://github.com/Hrazhan/sentiment). 
-It contains 1185 tweets labeled with positive, negative, and neutral sentiments.
+# Central Kurdish Sentiment (sentiment)
 
-[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+1,185 Central Kurdish tweets labelled for sentiment — the dataset behind the
+transfer-learning sentiment project, with integer labels `-1` (negative),
+`0` (neutral) and `1` (positive).
+
+## At a glance
+
+| | |
+|---|---|
+| Rows | 1,185 — `train` 948 / `test` 237 |
+| Columns | `tweet` (string), `sentiment` (int64) |
+| Parquet on disk | 0.10 MB |
+| Language | Central Kurdish / Sorani (ckb) |
+
+## Label distribution
+
+Measured on the full `train` split (948 rows):
+
+| Label | Meaning | Rows |
+|---|---|---|
+| `-1` | negative | 510 (53.8%) |
+| `1` | positive | 239 (25.2%) |
+| `0` | neutral | 199 (21.0%) |
+
+Tweet length in `train`: mean 77.7 characters, max 150.
+
+The class balance is skewed towards negative, so report macro-F1 alongside accuracy
+and consider class weights when fine-tuning.
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("razhan/sentiment")
+print(ds["train"][0])   # {'tweet': '...', 'sentiment': 1}
+```
+
+## Provenance
+
+Compiled from the [transfer learning project for sentiment
+analysis](https://github.com/Hrazhan/sentiment); the associated paper is
+[arXiv:2304.04703](https://arxiv.org/abs/2304.04703).
+
+## Known limitations
+
+- Very small: fine-tuning signal only, not a benchmark you should read much into.
+- Tweets are short and informal (dialect spellings, Latin-script inserts, mentions).
+- The three-way label set is unbalanced; `neutral` is the smallest class.
+
+## Related
+
+- `razhan/kurdish_tweets` — a larger 37,200-tweet sentiment set (positive /
+  negative / neutral) covering both Sorani and Kurmanji.