Keras Recommenders: reliable, state-of-the-art recommendations for ranking and retrieval

Building a recommendation system that is high-quality, high-performance, and hallucination-free can be a challenge. In this video, Yufeng Guo introduces Keras Recommenders (KerasRS), a library designed to help developers build reliable ranking and retrieval models with ease.

We’ll walk through a complete code example using the MovieLens dataset to build a Sequential Retrieval model. Using a Gated Recurrent Unit (GRU) to analyze a user's watch history, we will predict exactly which movie they are likely to watch next.

Because KerasRS is built on Keras 3, this workflow is compatible with your choice of backend: TensorFlow, JAX, or PyTorch.

In this video, you will learn:
- What Keras Recommenders is and why it’s useful
- How to prepare sequential data (using the "snake" method) for training
- How to build a two-tower architecture with a query tower (GRU) and candidate tower
- How to use the BruteForceRetrieval layer for accurate predictions

Resources:
Build and train a recommender system in 10 minutes using Keras and JAX → https://goo.gle/44VVj1J
Keras Recommenders Documentation → https://goo.gle/495DAG9
Check out the Code Example → https://goo.gle/3KBM140

Chapters:
0:00 - Introduction: LLMs vs. Keras Recommenders
0:40 - What is KerasRS?
2:09 - Installation & Setup
3:04 - Sequential Retrieval & GRU Explained
4:30 - Preparing the MovieLens Dataset
6:35 - Data Batching & Structure
7:17 - Building the Two-Tower Model
8:14 - Making Movie Predictions
8:28 - Conclusion & Next Steps

#Keras #MachineLearning #Recommenders

Subscribe to Google for Developers → https://goo.gle/developers

Speaker: Yufeng Guo
Products Mentioned: Google AI