How to implement a recommender system

These days, when it comes to capturing and holding an audience’s attention you must deliver the most relevant content possible. Whether searching for jobs or something more important, like looking at cat memes, your reader expects to see content relevant to them – personalized content. However, getting your users to optimal digital consumption nirvana (coin new term – check!) can be unexpectedly difficult. While it isn’t rocket science, it does involve some brain power…

Companies like Netflix and Pinterest have tackled this issue and pioneered personalization for their audiences. Through strategic integration of content and product recommendations systems, these two digital powerhouses have seen measurable ROI. Netflix says that its personalization platform saves the company more than $1B annually, and Pinterest estimates that more than 40 percent of user engagement is powered by its Related Pins recommender systems.

So what about the rest of us, who don’t have the same kinds of resources? Can we do recommender systems too? Yes, and here’s how. Read on to learn how two commonly used algorithms – matrix factorization and a bipartite graph—can be used to deliver personalization in an application.

Matrix factorization

Perhaps the most common type of recommender system algorithm is matrix factorization. The idea behind matrix factorization is to break a user-item feature matrix into a product of matrices, which end up becoming “user latent” features and “item latent” features. Latent features are hidden features that are derived from observed features using matrix factorization.

No Comments, Be The First!

Your email address will not be published.