Sentiment Analysis App

This is a production-ready Sentiment Analysis application in which I used Sentiment 140 dataset from Kaggle which contains 1.6 million tweets labeled by positive or negative.

• Cleaned the data by removing stopwords
• Used NLTK library with ( wordnet , omw-1.4 ) which gave synonyms , antonyms and multilingual support to be used for Lemmatization
• Applied Bigrams in Tfidf which gave huge bump in accuracy of the model because meaning in language often comes from word pairs, not single words
• Used Lemmatization over PorterStemmer to produce real, meaningful words, while stemming often produces crude, sometimes incorrect word fragments.
• Tuned Logistic Regression model by controling regularisation

Project Structure

• backend/: Contains the FastAPI application and training script.
• frontend/: Contains the static frontend files.
• requirements.txt: Python dependencies.
• render.yaml: Configuration for deploying to Render.

Setup & Installation

1. Install Dependencies:

   pip install -r requirements.txt

2. Train the Model: Before running the app, you need to generate the model and vectorizer artifacts. Run the training script (ensure the dataset training.1600000.processed.noemoticon.csv is in the root directory):

   cd backend
   python train.py
   cd ..

   This will create model.pkl and vectorizer.pkl in the backend/ directory.

3. Run the Application:

   uvicorn backend.main:app --reload

   The app will be available at http://127.0.0.1:8000.

Deployment on Render

1. Push to GitHub: Commit your code (including model.pkl and vectorizer.pkl if they are not too large, otherwise use Git LFS or train as part of build if dataset is small) to a GitHub repository. Note: Since the dataset is large, it is recommended to train locally and commit the .pkl files.

2. Create Web Service:

   • Connect your GitHub repo to Render.
   • Select "Web Service".
   • Render should automatically detect render.yaml.
   • If not, use the following settings:
     • Build Command: pip install -r requirements.txt
     • Start Command: uvicorn backend.main:app --host 0.0.0.0 --port $PORT

3. Access: Your app will be live at your Render URL.

Future Enhancements

• I will build a version 2 of this model by using BERT ( Bidirectional Encoder Representations from Transformers )
• It will give my model the abilty of reading the sentence BOTH ways
• Making it understand true meaning of the words by reading the sentence from left->right and right->left
• It would also be able to handle conditions like Misspellings , Slang , Rare words

👨‍💻 Author

Created with passion ⚡️ by Shrish Mishra

---

⭐ If you find this project useful, please consider giving it a star!