Google Introduces TensorFlow Recommenders, ‘Helping Users Find What They Love’

Original article was published by Synced on Artificial Intelligence on Medium

Google Introduces TensorFlow Recommenders, ‘Helping Users Find What They Love’

From the morning news stories you read to the late-night Netflix TV binges that consume you, recommender systems probably play a larger role in your everyday life than you might imagine. Companies like Google, Amazon, Spotify, YouTube and Netflix leverage this AI-powered technology to predict a user’s preferences and then suggest the most personalized “Guess You Like” content, which could take the form for example of ads or a playlist. The trend seems unstoppable, and the recommendation engine market is forecast to grow by US$3.57 billion from 2020–2024, a CAGR of almost 30 percent.

Google is one of the leading companies in recommender system research, development and deployment, and has been utilizing deep learning techniques such as multi-task learning, reinforcement learning and better user representations and fairness objectives to make its recommendations more personalized and effective. A group of researchers from Google Brain recently introduced a new open-sourced TensorFlow package, TensorFlow Recommenders (TFRS) designed to simplify the process of building, evaluating, and serving sophisticated recommender models.

Based on TensorFlow 2.x and Keras, TensorFlow Recommenders has the following features:

  • Build and evaluate flexible candidate nomination models
  • Freely incorporate item, user, and context information into recommendation models
  • Train multi-task models that jointly optimize multiple recommendation objectives
  • Efficiently serve the resulting models using TensorFlow Serving

The features are flexible and easy to use, and the modular design enables users to easily customize individual layers and metrics while maintaining a cohesive whole with individual components working well together.

The TensorFlow Package default settings aim to be sensible, with common tasks intuitive and straightforward to implement. More complex or custom recommendation tasks are also possible, and researchers say the goal is to make TFRS “an evolving platform, flexible enough for conducting academic research and highly scalable for building web-scale recommender systems.”

Google AI lead Jeff Dean promoted the new TensorFlow Recommenders in a tweet, suggesting developers have a look:

TensorFlow Recommenders is now open-sourced on GitHub. Google Brain researchers plan to expand the capabilities for multi-task learning, feature cross modelling, self-supervised learning and state-of-the-art efficient approximate nearest neighbours computation. Additional information and an illustrative use-case are available on the TensorFlow blog.