Update README.md

README.md

@@ -1,3 +1,103 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+language: en
+tags:
+  - text-classification
+  - sentiment-analysis
+  - imdb
+  - transformers
+  - distilbert-base-uncased
+---
+
+# IMDb Sentiment Dataset (8k Training Samples)
+
+This dataset is a **balanced subset** of the IMDb movie review sentiment dataset.  
+It is designed for **binary sentiment classification** experiments.  
+
+The dataset has been **cleaned to remove HTML tags** using BeautifulSoup and split into **training, validation, and test sets**.
+
+---
+
+## Dataset Overview
+
+- **Source:** IMDb sentiment dataset ([stanfordnlp/imdb](https://huggingface.co/datasets/stanfordnlp/imdb))  
+- **Task:** Binary sentiment classification  
+- **Labels:**
+  - 0 → Negative
+  - 1 → Positive
+
+---
+
+## Dataset Splits
+
+| Split      | Reviews | Positive | Negative |
+|------------|--------:|---------:|---------:|
+| Train      | 8,000   | 4,000    | 4,000    |
+| Validation | 2,000   | 1,000    | 1,000    |
+| Test       | 2,000   | 1,000    | 1,000    |
+
+---
+
+## Preprocessing
+
+- HTML tags removed using BeautifulSoup 
+- Cleaned text stored in `cleaned_text` column  
+- Original labels preserved in `label` column  
+- Data shuffled and balanced across classes  
+
+---
+
+## Validation Set Creation
+
+- The **training set (8k samples)** was sampled first, balanced across classes  
+- The **validation set (2k samples)** was sampled from remaining reviews, ensuring **no overlap** with training  
+- The **test set (2k samples)** was sampled from the remaining reviews after train & validation selection  
+- Each split was **shuffled** and **index reset** to avoid ordering bias  
+
+---
+
+## Files in This Dataset
+
+| File Name                        | Description |
+|---------------------------------|------------|
+| `imdb_cleaned_train_8000.csv`    | Training data (8k samples, cleaned) |
+| `imdb_cleaned_val_2000.csv`      | Validation data (2k samples, cleaned) |
+| `imdb_cleaned_test_2000.csv`     | Test data (2k samples, cleaned) |
+| `imdb_train_8000.csv`            | Original training subset (before cleaning) |
+| `imdb_val_2000.csv`              | Original validation subset (before cleaning) |
+| `imdb_test_2000.csv`             | Original test subset (before cleaning) |
+
+---
+
+## Intended Uses
+
+- Training sentiment analysis models (e.g., DistilBERT, BERT, or other Transformers)  
+- Benchmarking binary text classification tasks  
+- Educational NLP experiments  
+
+---
+
+## Limitations
+
+- Small subset (**8k training samples**) — may not generalize to all domains  
+- Only English movie reviews  
+- Long reviews may be truncated if tokenized for models with limited input length  
+
+---
+
+## References
+
+- Original IMDb dataset: [stanfordnlp/imdb](https://huggingface.co/datasets/stanfordnlp/imdb)  
+- Cleaning method: BeautifulSoup for HTML tag removal  
+- Example model trained on this dataset: DistilBERT IMDb Sentiment Analysis  
+
+---
+
+## Environment / Framework
+
+The dataset was prepared using the following Python libraries:
+
+- Python: 3.10  
+- Pandas: 1.6.1  
+- BeautifulSoup: 4.12.2  
+- Datasets: 4.10.1