Improve dataset card: document schema, measured stats and limitations

README.md

@@ -43,158 +43,71 @@ dataset_info:
       num_examples: 24692
 ---
 
-# Dataset Card for IMDB Kurdish
+# IMDB Kurdish (imdb_ckb)
 
-## Table of Contents
-- [Table of Contents](#table-of-contents)
-- [Dataset Description](#dataset-description)
-  - [Dataset Summary](#dataset-summary)
-  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
-  - [Languages](#languages)
-- [Dataset Structure](#dataset-structure)
-  - [Data Instances](#data-instances)
-  - [Data Fields](#data-fields)
-  - [Data Splits](#data-splits)
-- [Dataset Creation](#dataset-creation)
-  - [Curation Rationale](#curation-rationale)
-  - [Source Data](#source-data)
-  - [Annotations](#annotations)
-  - [Personal and Sensitive Information](#personal-and-sensitive-information)
-- [Considerations for Using the Data](#considerations-for-using-the-data)
-  - [Social Impact of Dataset](#social-impact-of-dataset)
-  - [Discussion of Biases](#discussion-of-biases)
-  - [Other Known Limitations](#other-known-limitations)
-- [Additional Information](#additional-information)
-  - [Dataset Curators](#dataset-curators)
-  - [Licensing Information](#licensing-information)
-  - [Citation Information](#citation-information)
-  - [Contributions](#contributions)
+Central Kurdish (Sorani) movie-review sentiment: 49,595 reviews labelled positive
+or negative, translated from the Stanford IMDB review set. Balanced labels, so a
+0.50 accuracy baseline is meaningless — report F1.
 
-## Dataset Description
+## At a glance
 
-- **Homepage:** [http://ai.stanford.edu/~amaas/data/sentiment/](http://ai.stanford.edu/~amaas/data/sentiment/)
-- **Repository:** [https://github.com/Hrazhan/IMDB_Kurdish/](https://github.com/Hrazhan/IMDB_Kurdish/)
-- **Point of Contact:** [Razhan Hameed](https://twitter.com/RazhanHameed)
-- **Paper:**
-- **Leaderboard:**
+| | |
+|---|---|
+| Rows | 49,595 — `train` 24,903 / `test` 24,692 |
+| Columns | `text` (string), `label` (`0` = negative, `1` = positive) |
+| Files | `train.csv` (59.4 MB), `test.csv` (57.5 MB) — CSV, not parquet |
+| Language | Central Kurdish / Sorani (ckb) |
 
-### Dataset Summary
+## Label balance
 
-Central Kurdish translation of the famous IMDB movie reviews dataset.
+Counted over the full CSVs:
 
-The dataset contains 50K highly polar movie reviews, divided into two equal classes of positive and negative reviews. We can perform binary sentiment classification using this dataset. 
-The availability of datasets in Kurdish, such as the IMDB movie reviews dataset, can help researchers and developers train and evaluate machine learning models for Kurdish language processing.
-However, it is important to note that machine learning algorithms can only be as accurate as the data they are trained on (in this case the quality of the translation), so the quality and relevance of the dataset will affect the performance of the resulting model.
-For more information about the dataset, please go through the following link: http://ai.stanford.edu/~amaas/data/sentiment/
+| Split | Positive (`1`) | Negative (`0`) |
+|---|---|---|
+| `train` | 12,484 | 12,445 |
+| `test` | 12,423 | 12,291 |
 
-P.S. This dataset is translated with Google Translator.
+## Parsing warning
 
-### Supported Tasks and Leaderboards
+The review text contains **embedded newlines**, so the CSVs have far more physical
+lines than records: `train.csv` has ~49,858 lines for 24,903 rows, `test.csv`
+~49,429 lines for 24,692 rows. Use a real CSV reader (`csv` module or
+`load_dataset`) — never `line.split(",")`.
 
-[More Information Needed]
+```python
+from datasets import load_dataset
 
-### Languages
-
-Central Kurdish
-
-## Dataset Structure
+ds = load_dataset("razhan/imdb_ckb")
+print(ds["train"][0])
+# {'text': 'فیلمێکی زۆر باش، کە سەرنج دەخاتە سەر پرسێکی زۆر گرنگ. ...', 'label': 1}
+```
 
-### Data Instances
+## Sample rows
 
-An example of 'train' looks as follows.
 ```
-{
-    "label": 0,
-    "text": ""فیلمێکی زۆر باش، کە سەرنج دەخاتە سەر پرسێکی زۆر گرنگ. نەخۆشی کحولی کۆرپەلە کەموکوڕییەکی زۆر جددی لە لەدایکبوونە کە بە تەواوی دەتوانرێت ڕێگری لێبکرێت. ئەگەر خێزانە زیاترەکان ئەم فیلمە ببینن، ڕەنگە منداڵی زیاتر وەک ئادەم کۆتاییان نەهاتبێت. جیمی سمیس لە یەکێک لە باشترین ڕۆڵەکانیدا نمایش دەکات تا ئێستا. ئەمە فیلمێکی نایاب و باشە کە خێزانێکی زۆر تایبەت لەبەرچاو دەگرێت و پێویستییەکی زۆر گرنگی هەیە. ئەمەش جیاواز نییە لە هەزاران خێزان کە ئەمڕۆ لە ئەمریکا هەن. منداڵان هەن کە لەگەڵ ئەم جیهانەدا خەبات دەکەن. بەڕاستی خاڵە گرنگەکە لێرەدا ئەوەیە کە دەکرا ڕێگری لە هەموو شتێک بکرێت. خەڵکی زیاتر دەبێ ئەم فیلمە ببینن و ئەوەی کە ه��یەتی بە جددی وەریبگرێت. بە باشی ئەنجام دراوە، بە پەیامی گرنگ، بە شێوەیەکی بەڕێزانە مامەڵەی لەگەڵ دەکرێت."
-}
+فیلمێکی زۆر باش، کە سەرنج دەخاتە سەر پرسێکی زۆر گرنگ. نەخۆشی کحولی کۆرپەلە کەموکوڕییەکی زۆر ...
+ئەم فیلمەم لە پۆلێکی ئایین بەکارهێنا کە فێری دەکردم. ماسییە زێڕینەکە بە خۆشحاڵیەوە لە قاپە ...
 ```
 
-### Data Fields
-
-
-plain_text
-
-    text: a string feature.
-    label: a classification label, with possible values including neg (0), pos (1).
-
-
-### Data Splits
-
-|   name   |train|test|
-|----------|----:|----:|
-|plain_text|24903|24692|
-
-## Dataset Creation
-
-### Curation Rationale
-
-[More Information Needed]
-
-### Source Data
-
-#### Initial Data Collection and Normalization
-
-[More Information Needed]
-
-#### Who are the source language producers?
-
-[More Information Needed]
-
-### Annotations
-
-#### Annotation process
+## Known limitations
 
-[More Information Needed]
+- **Translated data.** Reviews were rendered from English into Sorani, so idiom,
+  politeness and humour come from machine translation; cultural references remain
+  Western (film titles are transliterated).
+- Spam reviews that are part of the original IMDB set are included, exactly as in
+  the source: they are hard even for humans.
+- Domain is limited to film reviews — a linear probe trained here does not transfer
+  to news or social media.
+- The upstream licence applies (`other`); the IMDB source data is for
+  non-commercial research use only.
 
-#### Who are the annotators?
+## Provenance
 
-[More Information Needed]
-
-### Personal and Sensitive Information
-
-[More Information Needed]
-
-## Considerations for Using the Data
-
-### Social Impact of Dataset
-
-[More Information Needed]
-
-### Discussion of Biases
-
-[More Information Needed]
-
-### Other Known Limitations
-
-[More Information Needed]
-
-## Additional Information
-
-### Dataset Curators
-
-[More Information Needed]
-
-### Licensing Information
-
-[More Information Needed]
-
-### Citation Information
-
-```
-@InProceedings{maas-EtAl:2011:ACL-HLT2011,
-  author    = {Maas, Andrew L.  and  Daly, Raymond E.  and  Pham, Peter T.  and  Huang, Dan  and  Ng, Andrew Y.  and  Potts, Christopher},
-  title     = {Learning Word Vectors for Sentiment Analysis},
-  booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies},
-  month     = {June},
-  year      = {2011},
-  address   = {Portland, Oregon, USA},
-  publisher = {Association for Computational Linguistics},
-  pages     = {142--150},
-  url       = {http://www.aclweb.org/anthology/P11-1015}
-}
-
-```
+- Homepage: [ai.stanford.edu/~amaas/data/sentiment](http://ai.stanford.edu/~amaas/data/sentiment/)
+- Repository: [github.com/Hrazhan/IMDB_Kurdish](https://github.com/Hrazhan/IMDB_Kurdish/)
+- Contact: [Razhan Hameed](https://twitter.com/RazhanHameed)
 
-### Contributions
+## Related
 
-Thanks to [Razhan Hameed](https://twitter.com/RazhanHameed) for adding this dataset.
+- `razhan/kurdish_tweets`, `razhan/sentiment` — social-media sentiment instead of
+  long-form reviews.