This repo contains our project for the Wildberries Hackathon 2021.

Implement an algorithm of recommendation hints, which reduces the search for the desired products in the system.

• Natasha
• for Navec setup we use this model
• python-telegram-bot

• embeddings_prep.py is used for lemmatizing text words and mapping to their embedding representation. We use Natasha library.
• popularity.py is used for mapping words to their popularity.
• bot.py is used for launching bot.
• metrics.py is our small file to separate the cosine similarity metric we use from the rest of the project.

1. pip install requirements.txt in order to install all the required libraries.
2. Launch popularity.py to create popularity.pkl containing information about possible tags' relative 'popularity'.
3. Launch embeddings_prep.py to create embeddings.pkl containing information about possible tags' embeddings.
4. Launch bot.py and input your bot token to launch Telegram bot with your token which would allow you to get tags for you queries based on results of previous steps.

However, this repo also contains our pretrained version of the solution (the required for bot.py .pkl files) so it can be used out-of-box by installing required dependances, setuping Navec and simply running bot.py. Still, feel free to go into both embeddings_prep.py and popularity.py to try to tune some parameters.