Amazon, Spotify, Netflix, YouTube… most e-commerce companies and content providers incorporate recommender systems to their platforms. This way, they are able to recommend to each user products that they could potentially find interesting, increasing the user’s expertise and the company’s sales records.
These systems usually make recommendations to a user based on the preferences of similar users, an approach known as collaborative filtering. Over the last few years, the ML4DS group has designed several recommender systems which combine collaborative filtering models with methods based on similarities between users and/or content (avoiding the so-called “cold-start” problem). Furthermore, where needed, these systems have been adapted to Big-Data architectures over Spark.