Google Play Store’s app system is Now powered by DeepMind


DeepMind is widely regarded as the leading machine learning and artificial intelligence research lab. Under the alphabet, it collaborated with various Google teams over the years. DeepMind technology has been taken for the recommendation on the latest Play Store app.

Discovery is a central part of any App Store, and Google’s approach has both editorial and algorithmic. According to DeepMind, the Play Store supports “one of the largest recommendation systems in the world with billions of users every month”.

Google takes into account “previous user preferences” – downloads and installs – “to provide a rich, personalized experience.” DeepMind has collaborated with the Play team to “develop and improve the system that determines the relevance of the app in relation to the user.”

However, it requires nuances – both to understand what an app does and its relevance to a particular user. For example, for an avid sci-fi gamer, similar game recommendations may be of interest, but if a user installs a travel app, recommending a translation app may be more relevant than five more travel apps is.

At a high-level, recommendation system has three main models: a candidate generator, a ranker, and a model to optimize for multiple purposes. A blog post goes in-depth on each part today.

Google Play Store’s app system is Now powered by DeepMind

The candidate generator is a deep retrieval model that can analyze more than one million applications and retrieve the most appropriate ones. For each app, a reckoner, i.e. a user preference model, predicts user preferences along several dimensions.

Further, these predictions are the input for multi-purpose optimization models whose solutions give the most suitable candidate to the user.

DeepMind for the Google team threw some light into the collaboration process keeping in mind the daily dialogue.

Because the Play Store and DeepMind teams worked together and communicated on a daily basis, we took into account product requirements and constraints in the algorithm design, implementation, and final testing stages, resulting in a more successful product.


Please enter your comment!
Please enter your name here