Text Document Classification

This project focuses on classifying text documents using both supervised (Logistic Regression) and unsupervised (K-Means) learning techniques. It integrates machine learning with a Flask-based web application for deployment. The project also uses GitHub Actions for CI/CD and deploys the application using Docker and Heroku.

Key Features

1. Supervised Learning:

   • Logistic Regression for text classification.
   • Model training and evaluation, achieving an accuracy of 93%.

2. Unsupervised Learning:

   • K-Means clustering for document grouping.
   • Insights into natural groupings within the data.

3. Text Preprocessing:

   • Cleaning, tokenization, and vectorization of text.
   • Filtering rows based on word count to reduce variability.

4. Deployment:

   • Flask backend for serving predictions and clustering results.
   • Docker for containerization.
   • Heroku for seamless deployment with GitHub Actions for CI/CD.

Software and Tools Requirements

1. Github Account
2. Heroku Account
3. VS Code IDE
4. Git CLI

Project Workflow

1. Dataset:

   • Sourced from Kaggle.
   • Preprocessed by removing duplicates, unnecessary text, and filtering rows based on word count.

2. Supervised Learning:

   • Embeddings generated using Doc2Vec.
   • Trained a Logistic Regression model using an 80-20 train-test split.

3. Unsupervised Learning:

   • Embeddings created using BERT.
   • Dimensionality reduction applied with PCA.
   • Optimal clusters determined using the elbow method before training the K-Means model.

4. Visualization:

   • Word clouds to visualize the most frequent words in the dataset and each cluster.

5. Deployment:

   • Built a Flask API to handle predictions.
   • Automated the deployment process with GitHub Actions, Docker, and Heroku.

Setup and Usage

1. Clone the repository:

   git clone https://github.com/your-repo/document-classification.git
   cd document-classification
   

2. Create a new conda environment

   conda create -n ml python=3.11 -y
   

3. Activate the new conda environment

   conda activate ml
   

4. To install all the packages at once

   pip install -r requirements.txt
   

5. Run the Flask applicaiton locally:

   python app.py

Result

• The Logistic Regression model achieved an accuracy of 93%.
• K-Means clustering produced meaningful groupings, visualized using word clouds and scatter plots.

Future Improvement

- Enhance the clustering accuracy of the unsupervised model.
- Improve the web application's interface for better user experience.
- Scale the deployment for handling larger datasets and concurrent users.

Link and References

- https://www.geeksforgeeks.org/create-scatter-charts-in-matplotlib-using-flask/
- https://www.geeksforgeeks.org/how-to-add-graphs-to-flask-apps/
- https://medium.com/@danielafrimi/text-clustering-using-nlp-techniques-c2e6b08b6e95
- https://medium.com/@mishra.thedeepak/doc2vec-simple-implementation-example-df2afbbfbad5
- https://www.geeksforgeeks.org/doc2vec-in-nlp/
- https://www.kaggle.com/code/alhsan/word-statistics-effect-accuracy-0-973
- https://www.youtube.com/watch?v=MJ1vWb1rGwM&list=PLZoTAELRMXVMdvxeSuliQZcRLu3WCYVim

Video Link

- https://drive.google.com/file/d/1A_5T66xGOlWuW3Rf1-7VcifgA3ESNY_D/view?usp=drive_link