Improve dataset card: Add paper, code links, task category, and sample usage

This PR improves the dataset card for MahaParaphrase by:
- Adding the relevant paper link: https://huggingface.co/papers/2508.17444
- Including a link to the GitHub repository: https://github.com/l3cube-pune/MarathiNLP
- Specifying the `text-classification` task category in the metadata for better discoverability.
- Adding a "Sample Usage" section with installation instructions and a link to the Colab demo.
- Updating the citation to reflect the specific paper introducing this dataset.

README.md

@@ -1,17 +1,23 @@
 ---
-license: cc-by-4.0
 language:
 - mr
+license: cc-by-4.0
+size_categories:
+- 1K<n<10K
+pretty_name: MahaParaphrase
 tags:
 - paraphrase detection
 - Marathi NLP
 - Marathi paraphrase
-pretty_name: MahaParaphrase
-size_categories:
-- 1K<n<10K
+task_categories:
+- text-classification
 ---
+
 # L3Cube-MahaParaphrase Dataset
 
+Paper: [MahaParaphrase: A Marathi Paraphrase Detection Corpus and BERT-based Models](https://huggingface.co/papers/2508.17444)
+Code: https://github.com/l3cube-pune/MarathiNLP
+
 ## Overview:
 The **L3Cube-MahaParaphrase Dataset** is a Marathi paraphrase detection corpus.It is a high-quality, human-annotated corpus specifically designed for **Marathi**, a low-resource Indic language. It contains 8,000 sentence pairs labeled as either **Paraphrase (P)** or **Non-paraphrase (NP)**. This dataset is useful for tasks like paraphrase detection, semantic similarity, and data augmentation, as well as improving NLP models for low-resource languages.
 
@@ -38,13 +44,23 @@ The dataset is ideal for training and evaluating NLP models for:
 ## Model Benchmarks:
 Standard transformer-based models like **BERT** have been evaluated on this dataset, providing a performance baseline for future research.
 
+## Sample Usage
+
+This dataset is part of the `mahaNLP` library. You can install it via pip:
+
+```bash
+pip install mahaNLP
+```
+
+For usage examples, please refer to the [L3Cube-MahaNLP Colab demo](https://colab.research.google.com/drive/1POx3Bi1cML6-s3Z3u8g8VpqzpoYCyv2q).
+
 ## Citation:
 If you use this dataset, please cite the original work as follows:
-```
-@article{joshi2022l3cube,
-  title={L3cube-mahanlp: Marathi natural language processing datasets, models, and library},
+```bibtex
+@article{joshi2025mahaparaphrase,
+  title={MahaParaphrase: A Marathi Paraphrase Detection Corpus and BERT-based Models},
   author={Joshi, Raviraj},
-  journal={arXiv preprint arXiv:2205.14728},
-  year={2022}
+  journal={arXiv preprint arXiv:2508.17444},
+  year={2025}
 }
-```
+```