SentimentSeer-LSTM

SentimentSeer-LSTM is a repository featuring a sentiment analysis project that leverages Long Short-Term Memory (LSTM) networks. This notebook demonstrates the entire workflow of preparing text data, building an LSTM-based sentiment classification model, training it, and evaluating its performance.

Table of Contents

1. Overview
2. Getting Started
3. Data Preparation
4. Model Architecture
5. Training the Model
6. Model Evaluation
7. Usage
8. Contributing
9. License

Overview

SentimentSeer-LSTM provides a comprehensive implementation of sentiment analysis using LSTM networks. The primary components of this project include:

• Data Preparation: Includes downloading and preprocessing text data.
• Model Architecture: Describes the LSTM network used for sentiment classification.
• Training: Outlines the process of training the LSTM model on the preprocessed data.
• Evaluation: Details how to evaluate the model’s performance on test data.
• Inference: Shows how to use the trained model for making predictions on new text data.

Getting Started

To run the notebook, ensure that you have the following dependencies installed:

• Python (>= 3.6)
• PyTorch (>= 1.0)
• NumPy
• Collections
• String

Install the required Python packages using pip:

pip install numpy torch

Data Preparation

The dataset used consists of movie reviews and their associated sentiment labels. Data is first downloaded and then preprocessed, which includes:

• Reading Text Files: The reviews and labels are loaded from text files.
• Text Normalization: Converting all text to lowercase and removing punctuation.
• Tokenization: Splitting text into words and converting words to integer tokens.
• Padding: Ensuring that all sequences are of uniform length for model input.

Model Architecture

The sentiment analysis model is built using Long Short-Term Memory (LSTM) networks. The model architecture consists of:

• Embedding Layer: Transforms word indices into dense vectors of fixed size.
• LSTM Layer: Processes the sequences to capture temporal dependencies.
• Dropout Layer: Regularizes the model to prevent overfitting.
• Fully Connected Layer: Maps LSTM outputs to sentiment classes.
• Sigmoid Activation: Produces the final probability of the review being positive.

Training the Model

The model is trained using the following steps:

1. Data Loading: Data is split into training, validation, and test sets. Dataloaders are used to handle batching and shuffling of the training data.
2. Loss Function: Binary Cross Entropy Loss is used to measure the error in predictions.
3. Optimizer: Adam optimizer is used for updating the model’s parameters.
4. Epochs: The model is trained over a set number of epochs, with validation loss being monitored to prevent overfitting.

Trained Model -

• You can find trained model on HuggingFace - https://huggingface.co/joe-xhedi/SentimentSeer_LSTM/blob/main/sentiment_analysis_model.pth

Model Evaluation

After training, the model is evaluated on a separate test set to determine its performance. Key metrics include:

• Test Loss: Average loss on the test data => 0.477
• Test Accuracy: Proportion of correctly classified reviews. => 0.803 (trained for 4 epochs)

The evaluation helps in understanding how well the model generalizes to unseen data.

Usage

To use the trained model for predicting the sentiment of new reviews:

1. Tokenize and Pad Review: Convert the review into tokenized integers and pad it to the required sequence length.
2. Predict: Pass the processed review through the model to get the sentiment prediction.

Examples of how to use the model for prediction are provided in the notebook.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request with your changes. Ensure that your code adheres to the existing style and includes relevant tests.

License

This project is licensed under the MIT License - see the LICENSE file for details.