Alphabet’s DeepMind development of artificial intelligence and machine learning model architecture has dramatically improved the discovery system in the Google Play store, according to Google. In a blog post this morning, DeepMind detailed a partnership to strengthen the recommendation engine to support the Google Play Store, which has more than 2 billion Android users a month. It claims that, as a result, application recommendations are more personalized than before.
Notably, this isn’t the first time the DeepMind team has donated its expertise to Google’s Android business. The UK-based subsidiary created the Device Learning System to improve Android battery performance, and its WaveNet system is used to generate voice and now available to Google Assistant users. Google bought DeepMind in January 2014 for $400 million.
As DeepMind notes, Google Play’s recommendation system consists of three main models: candidate generators, rerankers, and AI models that can be optimized for multiple targets. Candidate generators can analyze more than a million applications and retrieve the most appropriate applications, while rerank programs can predict user preferences across multiple dimensions. The forecast results are used as input to the above optimization model, and the solution of the optimization model provides the most suitable candidate for the user.