Spotify leverages graph-based models to personalize audiobooks and podcasts. They use a graph structure to capture content relationships and user preferences. The system successfully overcame the lack of Audiobook interaction data and achieved significant increases in audiobook stream rates. The Unified Model architecture provides a foundation layer for representation learning and a dynamic layer for user representation learning, leading to improved recommendations.
Table of contents
Graph-based personalization for AudiobooksA unified model for personalizationConclusions37 Impressions