daheer / e-commerce-recommender

Recommendation system that searches similar items

E-Commmerce-Recommender

E-Commerce-Recommender is a recommendation system that aims to help users find similar items to the ones they bought from a retail store.
It does this using two components. A CLIP and Dot NN component. The CLIP part uses pretrained embeddings from the CLIP Text Model and the Dot NN part uses trained-from-scratch embeddings.

Model Architecture (Dot NN)

Built Using

• Python
• PyTorch
• Hugging Face
• FastAI
• CLIP
• FastAPI
• Others

Prerequisites and Installation

• Python

python -m pip install -r requirements.txt

• FastAPI

uvicorn main:app

Project Structure

│   main.py
│   train.ipynb
│   utils.py    
|
├── variables
│   ├──all_items.pt
│   ├──df.pt
|   └──weights.pt
|
└── images
    └─── model_architecture.png

Usage

Running the code mentioned above should launch a server in your localhost with some port. Head over to "bought" page and supply an index. Example:

localhost:8000/bought/1234

Demo

References

• Learning Transferable Visual Models From Natural Language Supervision

Contact

Dahir Ibrahim (Deedax Inc) - http://instagram.com/deedax_inc
Email - [email protected]
YouTube - https://www.youtube.com/@deedaxinc.3389
Twitter - https://twitter.com/DeedaxInc
Project Link - https://github.com/Daheer/E-Commerce-Recommender