juwume / reviews-parser Goto Github PK

This our final project in BMSTU.

1. Smyslov Maxim (Back-end)
2. Bogdanova Valeria (Front-end)

This app works as a web service, which parses marketplaces, then classifies comments as negative or positive and finaly sends the results in Telegram bot

• Telebot
• Flask
• Aiohttp
• TF-IDF as text transformer
• XGBoost as tonality classifier
• mongoengine as ORM
• MongoDB on parser side
• PostgreSQL on bot side

For classification was taken a dataset of human labeled comments from marketplaces. This dataset contains comments from cat food products and each comment is marked as positive or negative.

Precicion = 0.9260379462934971
Recall = 0.9390572924932551
Roc auc = 0.9466007732268533

1. Fill in the credentials in the .env file (there is an example)
2. docker-compose build
3. docker-compose up
4. Enjoy