Reach for the Top: How Spotify Built Shortcuts in Just Six Months

This post details the engineering behind Spotify's 'Shortcuts' feature on the Home screen. It covers the initial analysis of user listening habits to understand the opportunity, the development of recommendation heuristics (including decay functions and ranking by play frequency/duration), and the subsequent creation of a neural network-based ML model using TensorFlow. The ML model architecture, which learns event embeddings and scores entities, achieved a 26.7% relative improvement over the heuristic model offline. The post also discusses the validation process through A/B testing and the correlation of offline metrics (like NDCG) with online performance.