TensorFlow Sentiment Analysis on Google Play Reviews using Recurrent Neural Network (RNN)

Project Description

This project trains a Recurrent Neural Network (RNN) to analyze user reviews from the Google Play Store and classify them as positive, neutral, or negative. Leveraging deep learning techniques, this project aims to provide insights into user sentiment, which can inform app development and marketing strategies.

Technologies and Libraries

This project leverages several essential Python libraries:

• TensorFlow & Keras: For constructing, training, and evaluating the RNN model.
• Pandas: For data manipulation and analysis.
• scikit-learn: For dataset splitting and metric evaluation.
• Matplotlib: For data visualization.

Table of Contents

• TensorFlow Sentiment Analysis on Google Play Reviews using Recurrent Neural Network (RNN)
  • Project Description
  • Technologies and Libraries
  • Table of Contents
  • Installation
  • Dataset
  • Model Architecture
  • Training
  • Evaluation
  • Results
  • Usage
  • Contributing
  • License

---

Installation

To run this project, you need Python 3 and the following dependencies:

• tensorflow
• keras
• pandas
• numpy
• sklearn
• matplotlib
• pathlib
• pickle (if you want to save tokenizer)

Install these dependencies with:

pip install tensorflow keras pandas numpy scikit-learn matplotlib pathlib pickle

Dataset

The project uses a dataset of Google Play Store user reviews, googleplaystore_user_reviews.csv, with a total of 64,295 entries. Only records with non-null reviews and sentiment labels are used for training, which results in around 37,400 labeled entries. The dataset includes:

• Review: Text of the review.
• Sentiment: Label indicating if the review is positive, neutral, or negative.

Model Architecture

The model is built using a Recurrent Neural Network (RNN) architecture, ideal for processing sequential data like text. Key layers and components include:

• Embedding Layer: For converting text into dense vector representations.
• LSTM/GRU Layers: To capture contextual information from the sequence of words in the review.
• Dense Output Layer: A fully connected layer with three neurons (for positive, neutral, and negative classes), activated by softmax.

Training

To train the model, the dataset is split into training and testing sets. During training:

• Text preprocessing (tokenization and padding) is applied to standardize input sequences.
• The model is optimized using categorical cross-entropy as the loss function and the Adam optimizer.
• Training metrics are monitored to ensure the model's accuracy and reduce overfitting.

You can checkout the training process by running the Jupyter notebook cells sequentially.

Evaluation

After training, the model is evaluated on the test set, with metrics like accuracy and F1 score used to measure its performance. Additionally, a confusion matrix may be included to visualize the prediction distribution.

Results

The model demonstrates a strong ability to classify sentiments accurately:

• Accuracy: The model achieves satisfactory accuracy on both training and test datasets.
• Visualizations: Training and validation loss/accuracy plots provide insights into the model's performance and convergence.

Usage

The notebook includes code to predict sentiment for new reviews, replacing the variable sample_review with any review text to classify its sentiment:

predicted_sentiment = model.predict(sample_review)
print(predicted_sentiment)

Contributing

Contributions are welcome! If you'd like to improve this project or extend its functionality, please follow these steps:

1. Fork the repository.
2. Create a new branch (git checkout -b feature-branch).
3. Commit your changes (git commit -am 'Add new feature').
4. Push to the branch (git push origin feature-branch).
5. Create a new Pull Request.

License

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