Create README.md

README.md

@@ -0,0 +1,98 @@
+---
+datasets:
+- nhull/tripadvisor-split-dataset-v2
+tags:
+- nlp
+- hotels
+- reviews
+- sentiment-analysis
+- rnn
+- deep-learning
+---
+# LSTM Sentiment Analysis Model
+
+This is a fine-tuned LSTM model trained for sentiment analysis on the TripAdvisor dataset. The model predicts sentiment scores on a scale of 1 to 5 based on review text.
+
+- **Base Model**: Custom one-layer LSTM
+- **Dataset**: `nhull/tripadvisor-split-dataset-v2`
+- **Use Case**: Sentiment classification for customer reviews to understand customer satisfaction.
+- **Output**: Sentiment labels (1–5)
+
+___
+
+## Model Details
+
+- **Embedding**: 100-dimensional pre-trained GloVe embeddings
+- **Learning Rate**: 3e-04
+- **Batch Size**: 64
+- **Epochs**: 20 (early stopping with patience = 3)
+- **Dropout**: 0.2
+- **Tokenizer**: Custom tokenizer (vocabulary size: 10,000)
+- **Framework**: TensorFlow/Keras
+
+___
+
+## Intended Use
+
+This model is designed to classify hotel reviews based on their sentiment. It assigns a star rating between 1 and 5 to a review, indicating the sentiment expressed in the review.
+
+___
+
+## Dataset
+
+The dataset used for training, validation, and testing is nhull/tripadvisor-split-dataset-v2. It consists of:
+
+- **Training Set**: 30,400 reviews
+- **Validation Set**: 1,600 reviews
+- **Test Set**: 8,000 reviews
+
+All splits are balanced across five sentiment labels.
+
+___
+
+### Test Performance
+
+| Metric     | Value  |
+|------------|--------|
+| Accuracy   | 0.6041 |
+| Precision  | 0.60   |
+| Recall     | 0.60   |
+| F1-Score   | 0.60   |
+
+#### Classification Report (Test Set)
+
+| Label | Precision | Recall | F1-Score | Support |
+|-------|-----------|--------|----------|---------|
+| 1     |   0.72    |  0.73  |   0.73   |  1600   |
+| 2     |   0.52    |  0.52  |   0.52   |  1600   |
+| 3     |   0.57    |  0.48  |   0.52   |  1600   |
+| 4     |   0.52    |  0.45  |   0.49   |  1600   |
+| 5     |   0.65    |  0.84  |   0.73   |  1600   |
+
+### Confusion Matrix (Test Set)
+
+| True \\ Predicted | 1    | 2    | 3    | 4    | 5    |
+|-------------------|------|------|------|------|------|
+| **1**             | 1168 | 1222 | 1286 | 1334 | 710  |
+| **2**             | 1200 | 832  | 1600 | 1648 | 1024 |
+| **3**             | 1011 | 1347 | 768  | 1459 | 835  |
+| **4**             | 1096 | 1432 | 1496 | 720  | 920  |
+| **5**             | 1155 | 1491 | 1555 | 1603 | 1344 |
+
+---
+
+## Files Included
+
+- **`correct_predictions.csv`**: Contains correctly classified reviews with their real and predicted labels.
+- **`misclassified_predictions.csv`**: Contains misclassified reviews with their real and predicted labels, along with the difference.
+- **`glove.6B.100d.txt`**: Pre-trained 100-dimensional GloVe embeddings used for initializing the embedding layer in the models.
+
+---
+
+## Limitations
+
+1. **Domain-Specific**: The model was trained on the TripAdvisor Sentiment Dataset, so it may not generalize to other types of reviews (e.g., Amazon, Yelp) or domains (e.g., tech product reviews) without further fine-tuning.
+2. **Subjectivity**: Sentiment annotations are subjective and may not fully represent every user's perception, especially for neutral or mixed reviews.
+3. **Performance**: The model's performance for mid-range sentiment labels (e.g., 2 and 3) is lower compared to extreme sentiment labels (1 and 5), as these tend to have more nuanced language.
+4. **Dependency on Pre-trained Embeddings**: The models rely on pre-trained GloVe embeddings, meaning their performance is closely tied to the quality and representativeness of these embeddings. Since GloVe embeddings were trained on a large, general corpus, they may not fully capture domain-specific nuances, such as specific phrasing or terms in hotel and restaurant reviews.
+5. **GRU Simplicity**: While the GRU model offers a simpler and more efficient architecture, it may not capture complex sequential patterns as effectively as more sophisticated models like BiLSTM or LSTM. This simplicity, however, contributes to its strong generalization performance on cross-domain data (e.g., Michelin dataset).