🎯 Recommendations Systems

The goal of this projet is to recommend books / articles to customers, integrate in a Node.js Mobile App.

Different types of recommendation systems

Content-Based Filtering: the theory

Collaborative Filtering: the theory

🗂️ Dataset

News Portal provided by Globo.com, one of the most popular media company in Brazil.

More information available on Gabriel Moreira's Github/paper

Notes on the data

In this project, we have IMPLICIT data, i.e. we don't have clear (explicit) feedback from the users about their preferences, like articles' rating.

Besides, there are additional drawbacks:

• We don't have full data about interactions: it would have been interesting to have at least the view duration of an article;
• We don't have the details of articles for confidentiality reasons: it's difficult to see "visually" if the recommendations are really relevant.

📜 Tasks

• ✔️ Perform Exploratory Data Analysis (EDA);
• ✔️ Try different RecSys model (by popularity, Content-Based, Collaborative Filtering);
• ✔️ Select the architecture to meet the business need;
• ✔️ Integrate in a Node.js Mobile App;
• ✔️ Deploy Content-based RecSys on Azure (Azure Functions, Azure Blob Storage).

💻 Dependencies

Pandas, sklearn, implicit library, Azure Functions, Azure Blob Storage, Github, VS Code

Targeted Serverless Architecture

🛠️ Tools (prerequisites)

• Fork of Bookshelf App on Github;
• Node.js, including npm (Node package manager);
• Azure Functions Core Tools;
• Android Studio.

📌 References

• Recommendations systems: Categories of RecSys
• Implicit / Explicit data :
• Cosine similarity : Sklearn pairwise metrics;
• Implicit library;
• Create Azure Functions with Visual Studio Code;
• Serverless : What is Serverless?

⏭️ Next steps...

• ☑️ Check if the articles have been seen more than once by an user;
• ☑️ Try dataset with explicit feedback;
• ☑️ Use scikit-surprise library;
• ☑️ what about hybrid and neural networks RecSys?