Item-based recommendation engine incl. custom techniques

A Python movie recommender system using cosine similarity, weighted rating, custom hyperparameter tuning etc.

1. Intro & required prerequisite

As a preface, this is neither a fundamental tutorial nor a foundational explanation of theory behind recommender systems. This article is about data preprocessing, some feature engineering, ML model development and the way I did it. It’s also assumed, that you’re more or less familiar with: Python, Pandas, Numpy and basic linear algebra. There will be also some parts of code in here.

So, what is actually a recommender system?

Inherently, recommendation system is a specific class of algorithms in Machine Learning, that offer the user “relevant” information. There are plenty of examples out there: Medium article suggestion, Netflix & movie recommendations, Google and their news feed etc. Basically, recommender systems are developed using bunch of algorithms and ML techniques, in order to suggest users “relevant” items, so that user can engage with them and create more traffic: reading more Medium articles, watching more movies on Netflix, reading more news on Google News etc. All the items are ranked based on their relevancy and individually calculated for every single user.