Paper
Deploying Semantic ID-based Generative Retrieval for Large-Scale Podcast Discovery at Spotify
Authors
Edoardo D'Amico, Marco De Nadai, Praveen Chandar, Divita Vohra, Shawn Lin, Max Lefarov, Paul Gigioli, Gustavo Penha, Ilya Kopysitsky, Ivo Joel Senese, Darren Mei, Francesco Fabbri, Oguz Semerci, Yu Zhao, Vincent Tang, Brian St. Thomas, Alexandra Ranieri, Matthew N. K. Smith, Aaron Bernkopf, Bryan Leung, Ghazal Fazelnia, Mark VanMiddlesworth, Timothy Christopher Heath, Petter Pehrson Skiden, Alice Y. Wang, Doug J. Cole, Andreas Damianou, Maya Hristakeva, Reid Wilbur, Tarun Chillara, Vladan Radosavljevic, Pooja Chitkara, Sainath Adapa, Juan Elenter, Bernd Huber, Jacqueline Wood, Saaketh Vedantam, Jan Stypka, Sandeep Ghael, Martin D. Gould, David Murgatroyd, Yves Raimond, Mounia Lalmas, Paul N. Bennett
Abstract
Podcast listening is often grounded in a set of favorite shows, while listener intent can evolve over time. This combination of stable preferences and changing intent motivates recommendation approaches that support both familiarity and exploration. Traditional recommender systems typically emphasize long-term interaction patterns, and are less explicitly designed to incorporate rich contextual signals or flexible, intent-aware discovery objectives. In this setting, models that can jointly reason over semantics, context, and user state offer a promising direction. Large Language Models (LLMs) provide strong semantic reasoning and contextual conditioning for discovery-oriented recommendation, but deploying them in production introduces challenges in catalog grounding, user-level personalization, and latency-critical serving. We address these challenges with GLIDE, a production-scale generative recommender for podcast discovery at Spotify. GLIDE formulates recommendation as an instruction-following task over a discretized catalog using Semantic IDs, enabling grounded generation over a large inventory. The model conditions on recent listening history and lightweight user context, while injecting long-term user embeddings as soft prompts to capture stable preferences under strict inference constraints. We evaluate GLIDE using offline retrieval metrics, human judgments, and LLM-based evaluation, and validate its impact through large-scale online A/B testing. Across experiments involving millions of users, GLIDE increases non-habitual podcast streaming on Spotify home surface by up to 5.4% and new-show discovery by up to 14.3%, while meeting production cost and latency constraints.
Metadata
Related papers
Vibe Coding XR: Accelerating AI + XR Prototyping with XR Blocks and Gemini
Ruofei Du, Benjamin Hersh, David Li, Nels Numan, Xun Qian, Yanhe Chen, Zhongy... • 2026-03-25
Comparing Developer and LLM Biases in Code Evaluation
Aditya Mittal, Ryan Shar, Zichu Wu, Shyam Agarwal, Tongshuang Wu, Chris Donah... • 2026-03-25
The Stochastic Gap: A Markovian Framework for Pre-Deployment Reliability and Oversight-Cost Auditing in Agentic Artificial Intelligence
Biplab Pal, Santanu Bhattacharya • 2026-03-25
Retrieval Improvements Do Not Guarantee Better Answers: A Study of RAG for AI Policy QA
Saahil Mathur, Ryan David Rittner, Vedant Ajit Thakur, Daniel Stuart Schiff, ... • 2026-03-25
MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination
Zhuo Li, Yupeng Zhang, Pengyu Cheng, Jiajun Song, Mengyu Zhou, Hao Li, Shujie... • 2026-03-25
Raw Data (Debug)
{
"raw_xml": "<entry>\n <id>http://arxiv.org/abs/2603.17540v1</id>\n <title>Deploying Semantic ID-based Generative Retrieval for Large-Scale Podcast Discovery at Spotify</title>\n <updated>2026-03-18T09:46:10Z</updated>\n <link href='https://arxiv.org/abs/2603.17540v1' rel='alternate' type='text/html'/>\n <link href='https://arxiv.org/pdf/2603.17540v1' rel='related' title='pdf' type='application/pdf'/>\n <summary>Podcast listening is often grounded in a set of favorite shows, while listener intent can evolve over time. This combination of stable preferences and changing intent motivates recommendation approaches that support both familiarity and exploration. Traditional recommender systems typically emphasize long-term interaction patterns, and are less explicitly designed to incorporate rich contextual signals or flexible, intent-aware discovery objectives. In this setting, models that can jointly reason over semantics, context, and user state offer a promising direction. Large Language Models (LLMs) provide strong semantic reasoning and contextual conditioning for discovery-oriented recommendation, but deploying them in production introduces challenges in catalog grounding, user-level personalization, and latency-critical serving.\n We address these challenges with GLIDE, a production-scale generative recommender for podcast discovery at Spotify. GLIDE formulates recommendation as an instruction-following task over a discretized catalog using Semantic IDs, enabling grounded generation over a large inventory. The model conditions on recent listening history and lightweight user context, while injecting long-term user embeddings as soft prompts to capture stable preferences under strict inference constraints. We evaluate GLIDE using offline retrieval metrics, human judgments, and LLM-based evaluation, and validate its impact through large-scale online A/B testing. Across experiments involving millions of users, GLIDE increases non-habitual podcast streaming on Spotify home surface by up to 5.4% and new-show discovery by up to 14.3%, while meeting production cost and latency constraints.</summary>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.IR'/>\n <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n <published>2026-03-18T09:46:10Z</published>\n <arxiv:primary_category term='cs.IR'/>\n <author>\n <name>Edoardo D'Amico</name>\n </author>\n <author>\n <name>Marco De Nadai</name>\n </author>\n <author>\n <name>Praveen Chandar</name>\n </author>\n <author>\n <name>Divita Vohra</name>\n </author>\n <author>\n <name>Shawn Lin</name>\n </author>\n <author>\n <name>Max Lefarov</name>\n </author>\n <author>\n <name>Paul Gigioli</name>\n </author>\n <author>\n <name>Gustavo Penha</name>\n </author>\n <author>\n <name>Ilya Kopysitsky</name>\n </author>\n <author>\n <name>Ivo Joel Senese</name>\n </author>\n <author>\n <name>Darren Mei</name>\n </author>\n <author>\n <name>Francesco Fabbri</name>\n </author>\n <author>\n <name>Oguz Semerci</name>\n </author>\n <author>\n <name>Yu Zhao</name>\n </author>\n <author>\n <name>Vincent Tang</name>\n </author>\n <author>\n <name>Brian St. Thomas</name>\n </author>\n <author>\n <name>Alexandra Ranieri</name>\n </author>\n <author>\n <name>Matthew N. K. Smith</name>\n </author>\n <author>\n <name>Aaron Bernkopf</name>\n </author>\n <author>\n <name>Bryan Leung</name>\n </author>\n <author>\n <name>Ghazal Fazelnia</name>\n </author>\n <author>\n <name>Mark VanMiddlesworth</name>\n </author>\n <author>\n <name>Timothy Christopher Heath</name>\n </author>\n <author>\n <name>Petter Pehrson Skiden</name>\n </author>\n <author>\n <name>Alice Y. Wang</name>\n </author>\n <author>\n <name>Doug J. Cole</name>\n </author>\n <author>\n <name>Andreas Damianou</name>\n </author>\n <author>\n <name>Maya Hristakeva</name>\n </author>\n <author>\n <name>Reid Wilbur</name>\n </author>\n <author>\n <name>Tarun Chillara</name>\n </author>\n <author>\n <name>Vladan Radosavljevic</name>\n </author>\n <author>\n <name>Pooja Chitkara</name>\n </author>\n <author>\n <name>Sainath Adapa</name>\n </author>\n <author>\n <name>Juan Elenter</name>\n </author>\n <author>\n <name>Bernd Huber</name>\n </author>\n <author>\n <name>Jacqueline Wood</name>\n </author>\n <author>\n <name>Saaketh Vedantam</name>\n </author>\n <author>\n <name>Jan Stypka</name>\n </author>\n <author>\n <name>Sandeep Ghael</name>\n </author>\n <author>\n <name>Martin D. Gould</name>\n </author>\n <author>\n <name>David Murgatroyd</name>\n </author>\n <author>\n <name>Yves Raimond</name>\n </author>\n <author>\n <name>Mounia Lalmas</name>\n </author>\n <author>\n <name>Paul N. Bennett</name>\n </author>\n </entry>"
}