Advanced machine learning helps Play Store users discover personalised apps

Over the past few years we've applied DeepMind's technology to Google products and infrastructure, with notable successes like reducing the amount of energy needed for Google Play Store to download apps for their mobile devices – the Play Store supports one of the largest recommendation systems in the world. While some are looking for specific apps, like Snapchat, others are browsing the store to discover what’s new and interesting. The Google Play discovery team strives to help users discover the most relevant apps and games by providing them with helpful app recommendations. To deliver a richer, personalised experience, apps are suggested according to past user preferences. This, however, requires nuance – both for understanding what an app does, and its relevance to a particular user. For example, to 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. The collection and use of these user preferences is governed by Google's privacy policies.

We started collaborating with the Play store to help develop and improve systems that determine the relevance of an app with respect to the user. In this post, we’ll explore some of the cutting-edge machine learning techniques we developed to achieve this. Today, Google Play’s recommendation system contains three main models: a candidate generator, a reranker, and a model to optimise for multiple objectives. The candidate generator is a deep retrieval model that can analyse more than a million apps and retrieve the most suitable ones. For each app, a reranker, i.e. a user preference model, predicts the user's preferences along multiple dimensions. Next these predictions are the input to a multi-objective optimisation model whose solution gives the most suitable candidates to the user.

Applied machine learning under real-world constraints

To improve how Google Play’s recommendation system learns users’ preferences, our first approach was to use an cooling Google’s data centres by up to 30%, boosted the value of Google’s our research collaboration with Waymo has helped improve the performance of its models, as well as the efficiency of training its neural networks.

Working at Google scale presents a unique set of research challenges, and the opportunity to take our breakthroughs beyond the lab to address global, complex challenges. If you’re interested in working on applying cutting edge research to real world problems, learn more about the team that led this project here.