Recommendations shape our reality. From the shows we watch and the products we buy to the news we consume and the people we date, recommendation engines play an integral role in our lives. They also play an integral role in your customers’ experiences, so it’s important to leverage the correct technique.

Realizing the importance of its own recommendation engine, Twitter recently outlined the mechanics of its recommendation algorithm at the end of March. To curate a personalized and relevant timeline from 500 million daily tweets, Twitter revealed that it relies on an ensemble of recommendation approaches. For example, when recommending tweets from profiles you don’t follow, Twitter looks to similar users, analyzing their engagements to find tweets that you may also like. When ranking tweets to recommend to users, however, Twitter employs deep learning and reinforcement learning by utilizing a multimillion-parameter neural network that is continuously trained to optimize positive engagement such as likes and retweets. Twitter also relies on clustering users into communities to deliver recommendations based on content similarity — for example, users who Twitter identifies as pop music fans are assigned to the pop music community, where they receive popular tweets in that category.

As the Twitter example shows, there are a variety of techniques available for recommendations. Business leaders must familiarize themselves with these techniques and find those that work for them. In our recently published report, Rev Your Recommendation Engines With New Techniques, we break down four recommendation methods:

  • Collaborative filtering relies on data from similar users to surface related content and products. For example, Twitter looks for users who also like tweets about chihuahuas and then will recommend other tweets with which chihuahua lovers engage.
  • Content-based filtering recommends content and products with similar attributes to those that the customer uses. For example, a social media platform may identify a user’s interest in basketball and then serve them posts with the latest in college basketball news.
  • Deep learning systems are a form of content-based filtering where techniques such as natural language processing and computer vision identify text, audio, image, and other unstructured attributes to deliver similar recommendations. For example, Indian music streaming platform Gaana can detect similarities in a song’s tempo to inform its recommendations.
  • Reinforcement learning systems deliver recommendations by experimenting and learning in a feedback loop. For example, a system using this method may recommend you posts about the latest pop culture drama, and if you respond positively by sharing the post, the system will continue dishing out celebrity gossip.

To learn more about the strengths and weaknesses of these systems and our recommendations on recommendations, read our report, Rev Your Recommendation Engines With New Techniques.