Thesis' Recommender System

• This repo contains only the source code of the recommender system which is a REST-ful API that allows calls from clients (my React front-end) to perform tasks like building the models (Incremental SVD and Item-based KNN), recommendations for users, related products for products.

Live demo: https://thesis-frontend.herokuapp.com (NOT WORKING)

⚠️ Note: The live demo can be a bit slow as start up since heroku put applications in sleep mode after a short period of inactive. Moreover, since the application is divided into three parts: ReactJS front-end, NodeJS back-end, Flask recommender system, each of them only "wakes up" when its functions are being used (especially for recommender system, it only starts when log-in as admin and use the recommender system), some functions may be slow at first.

• 🔴 Login as admin: username "admin", password "admin" => head to /recommender route to use the recommender system (train/test/save model, dataset) and other small features.
• 🔵 Login as usual user: username "A2GKMXRLI7KLFP", password "123" => explore the webiste functions like browsing/search/add-to-cart products, sort/filter products, etc.

Paper I published along with this project: https://github.com/tisu19021997/my-notebooks/tree/main/svd-xquad

Frameworks

• Web Stack:

  • Recommender System: Flask (this repo)
  • Front-end: ReactJS (https://github.com/tisu19021997/thesis-react)
  • Back-end: NodeJs and ExpressJS (https://github.com/tisu19021997/thesis-node)
  • Database: MongoDB

• Machine Learning:

  • Fast SVD: funk-svd
  • Collaborative Filtering: surprise-scikit
  • Data Analysis and Manipulation: pandas
  • Utilities: numpy

• Storage: model files are stored on Microsoft Azure storage (student free tier).

Folders and Files

• algo/: implementation of two core algorithms: Explicit Query Aspect Diversification (xQuAD) and the incremental version of Simon Funk's Singular Value Decomposition (SVD). Details about the methodology can be found in my other repository.
• lib/funk_svd/: the funk-svd library.
• helper/:
  • accuracy.py: some ranking metrics such as Average Recommendation Popularity (ARP), Average Popularity of Long-tail items (APLT), Average Coverage of Long-tail items (ACLT).
  • Others: some utils function, nothing important.
• wrapper/RecSys.py: the recommender system class. Built to be used on two algorithms: SVD for user recommendation and Item-based KNN for generating related products.