blanchonnicolas / ia_systemederecommandation

Project "My Content"

Context: Recommender System Goals: Solutions supporting user in article choice, by providing top 5 recommandations Repository of OpenClassrooms project 9' AI Engineer path

My Content

Our role is to participate to the conception and development of the recommender system:

• Item-Based (using article metadata and embeddings)
• Collaborative filtering - Model based
• Deploy solution on Cloud, using serverless platform (Azure function)

The project is using below dataset to train recommender systems: My Content dataset.

Finally, recommender system engine is callable using : Azure function called with user_id 72 - Http trigger

How to use

For Setup:

• Python and VSCode with Jupyter extension to read notebooks. For application access:
• Acces to web, through Streamlit-share app.

This repository is part of a 3-repos project :

• Main repo : Notebooks and Scripts : this repo
• Azure Function
• Mobile App

More details here :

• Presentation

• Streamlit App : Disabled

• Vidéo

Main Repo content

• Notebook 1 : datagenerator

  • Data Analysis: Visualisation, Modification
  • Unit Test

• Train Collaboraive Filtering model - script

  • Rating generation (Clicks_hint)
  • Creation of Scipy.sparse csr matrix (User/Article/Rating)
  • Training of Implicit ALS model (model.fit())
  • Joblib save using compression

• Compute CF Model Results - script

  • Metrics generation (Clicks_hint, Precision@k, MeanAverage@k)
  • Test_User_id loop to recomend using trained Implicit model, and compare prediction with real
  • Measure and store results

• Compute Item based Results - script

  • Metrics generation (Clicks_hint, Precision@k, MeanAverage@k)
  • Test_User_id loop to recomend using COSINE SIMILARITY from Embedding, and compare prediction with real
  • Measure and store results