ACM RecSys Challenge

RecSys Challenge 2026:
Conversational Music Recommendation

The Music Conversational Recommendation Challenge (Music-CRS) is a global research competition bridging Natural Language Processing and Recommender Systems. Participants build systems that engage in multi-turn conversations about music, recommend relevant tracks, and generate natural language responses.

The challenge uses TalkPlayData-Challenge, a large-scale synthetic multi-turn dialogue dataset grounded in real music listening histories, with pre-extracted multimodal track and user embeddings provided.

Sony SiriusXM Deezer
Final Results Analysis Register Closed Codabench Resources Terms and Conditions

Workshop Schedule

Friday, 2 October 2026
All times are local.

8:30–8:35Opening Remarks
8:35–9:25 From Language to Items: Building Conversational Recommendation Experiences at SpotifyKeynote 1 · Enrico Palumbo · Spotify · 40 min + 10 min Q&A

Abstract

The boundary between search and recommendation is dissolving, and at the center of that shift is a harder problem: connecting the open-endedness of natural language with the concreteness of items in a catalog. At Spotify, we've been exploring how to bridge this gap — building systems that can interpret complex, evolving user intent and translate it into precise, relevant retrieval. In this talk, I'll cover the generative retrieval models, agent architectures, multi-turn interaction, and data strategies behind this work, along with the challenges and open questions that come with treating recommendation as a language problem.

9:25–9:40 Overview and Analysis of the RecSys Challenge 2026: Conversational Music Recommendation Seungheon Doh, Sergio Oramas, Bruno Sguerra, Abhinav Bohra, Claudio Pomo, Francesco Barile [Paper]
9:40–10:00Paper Session 1: Auditing the Benchmark
Labels vs. requests and metric consistency
  • Auditing Alignment among Item Relevance, Goal Progress, and LLM-Judged Responses in Music-CRSTomoya Terai[Paper] [Code]
  • When the Label Ignores the Request: Auditing Policy-Selected Targets in Synthetic Conversational Music RecommendationSanjeev Suresh[Paper] [Code]
10:00–10:30Coffee Break
10:30–12:00Paper Session 2: Building the System
From multi-source retrieval to grounded responses
  • Talk Less, Rank Better: A Decoupled Pipeline for Conversational Music RecommendationAbdallah Alkhetiar, Luigi Inguaggiato, Nicolò Locatelli, Roberto Manea, Alessio Pizzini, David Ravelli, Gianmarco Schifone, Matteo Vitali, Andrea Zhang, Michael Benigni, Andrea Pisani, Maurizio Ferrari Dacrema[Paper] [Code]
  • A Practical Multi-Source Pipeline for Conversational Music Recommendation in the RecSys Challenge 2026Ryohei Wakatsuki[Paper] [Code]
  • A Multi-Source Retrieve-Fuse-Rerank System for Conversational Music RecommendationYouness Soussou, Loubna Mekouar, Youssef Iraqi[Paper] [Code]
  • Dialogue-Aware Conversational Music Recommendation via Multi-Modal Retrieval and Weighted RRFAditya Rai, Venkatesh Shukla, Sanskar Aggarwal, Amandeep Kaur[Paper] [Code]
  • Dialogue-Aware Music Recommendation via Fused Retrieval and Learned Ranking for the TalkPlayData ChallengeSuryaa Veerabathiran Seran[Paper] [Code]
  • MiniMaestro: Resource-Conscious Conversational Music Recommendation with a Single Open-Weight 8B ModelSimran Sundrani, Mohan Bhambhani[Paper] [Code]
  • Two Views, One Voice: Evidence-Grounded Conversational Music RecommendationSungwook Yoo, Sewook Yoo[Paper] [Code]
  • Picking is Not Ranking, and Explanation Quality Has Many Dimensions: Lessons for Conversational Music RecommendationMaxime Manderlier, Fabian Lecron[Paper] [Code]
  • Beyond Single-Signal Retrieval: A Graph-Aware Agentic Framework for Conversational Music RecommendationMarco Valentini, Bianca Di Bitetto, Gianmichele De Palma, Berardino Como, Roberta Russo, Francesco Maria Dicataldo, Francesco Salvatore Desiderato, Francesco Falcone, Chiara Mallamaci, Antonio Ferrara, Daniele Malitesta, Tommaso Di Noia, Fedelucio Narducci[Paper] [Code]
12:00–13:30Lunch Break
13:30–14:10 Netflix Prize Revisited: A Hard Problem Even After 20 YearsKeynote 2 · Domonkos Tikk · Taboola · 40 min Gábor Takács, István Pilászy, Bottyán András Németh, Balázs Hidasi, László Környei, Domonkos Tikk [Paper]
14:10–15:00Paper Session 3: Ceilings, Shift and Negative Results
Why gains do not transfer
  • Distribution-Robust Reranking for Conversational Music Recommendation: PoliBaJukebox at the ACM RecSys Challenge 2026Andrea Lops, Nicola Cipriani, Gabriele Colapinto, Marcantonio de Candia, Vito Di Bari, Mauro Foglia, Gledjan Meta, Davide Savoia, Antonio Ferrara, Daniele Malitesta, Tommaso Di Noia, Fedelucio Narducci[Paper] [Code]
  • Frozen Embeddings, LLM Retrieval Lanes, Critic-Guided Replies: Conversational Music Recommendation at an Information-Limited CeilingArtem Volgin, Ben-Orli Nathanson, Daniele Dal Grande, Alisa Milchevskaya, Adham Al Hossary, Rami Albatal[Paper] [Code]
  • Routing Discovery in Conversational Music RecommendationMadhav Patil, Shreyas Biradar, Divyansh Bhatia[Paper] [Code]
  • An Empirical Analysis of Retrieval and Evaluation in the RecSys Challenge 2026 Music-CRS TrackSiwon Lee, Young-June Choi[Paper] [Code]
  • State-Driven Retrieval and Learned Re-Ranking for Conversational Music RecommendationNidhin Pattaniyil, Semih Yagli, Tanwir Zaman[Paper] [Code]
15:00–15:30Coffee Break
15:30–16:30Poster Session
16:30–17:00Award Ceremony and Closing

Invited Speakers

Enrico Palumbo

Enrico Palumbo

Enrico Palumbo is a Senior Research Scientist at Spotify, previously at Amazon Alexa. His research focuses on improving Search and Recommendations through Generative AI, with a recent interest in agentic systems and generative recommendation. He has been a core contributor to the design and launch of AI products used by hundreds of millions of users, including Spotify's Agentic Search, Query Autocomplete, Conversational Agent, and Alexa's non-English models.

His work has resulted in several patents and publications in top-tier venues such as RecSys, KDD, WebConf, CIKM, and ESWA. He holds a PhD in Knowledge Graph Embeddings for Recommender Systems, which he carried out jointly between the Polytechnic University of Turin, EURECOM, and Links Foundation.

Istvan Pilaszy

Istvan Pilaszy

Istvan Pilaszy is a co-founder of Gravity R&D, a recommender systems company acquired by Taboola in 2022. He was a member of Gravity’s team in the Netflix Prize competition, which eventually joined The Ensemble, tying for first place in the final competition. He holds a PhD in Computer Science from the Budapest University of Technology and Economics, where his research focused on factorization-based large-scale recommendation algorithms.

At Gravity R&D, Istvan was responsible for the core recommendation engine, ensuring its scalability, efficiency, code quality, and reliability. He developed new recommendation algorithms, brought them into production, and scaled them to real-world workloads. Gravity’s recommendation engine lives on at Taboola and continues to evolve. Since the acquisition, Istvan has focused on real-time bidding and adapting the engine to the needs of a much broader range of use cases.

Domonkos Tikk

Domonkos Tikk

Domonkos Tikk is a co-founder of Gravity R&D, a recommender systems company acquired by Taboola in 2022, and currently leads Taboola’s R&D site in Hungary. He holds a PhD in computer science from the Budapest University of Technology and Economics. His research interests span recommender systems, machine learning, and text mining. He led Gravity’s team in the Netflix Prize competition, which eventually joined The Ensemble, tying for first place in the final competition.

Domonkos has been an active member of the ACM RecSys community for more than 15 years, serving in numerous organizational roles, including RecSys Challenge Co-Chair, Industry Co-Chair, and Program Co-Chair of RecSys 2019. He is also a member of the ACM RecSys Steering Committee. He has co-authored approximately 200 publications, which have received more than 13,000 citations. He is also a recipient of an Alexander von Humboldt Research Fellowship for experienced researchers.

Organizing Committee

Timeline

31 March 2026 Website published.
10 April 2026 Start of the RecSys Challenge; Train, Development, and Blind A datasets released.
17 April 2026 Submission system opens; leaderboard live for Blind A.
23 June 2026 Blind Dataset B released; Blind B final phase opens.
30 June 2026 End of the RecSys Challenge.
6 July 2026 Final leaderboard and winners announced; EasyChair opens for paper submissions.
9 July 2026 Code upload deadline for final predictions.
20 July
24 July 2026		 Paper submission deadline.
3 August
5 August 2026		 Paper acceptance notifications.
10 August 2026 Camera-ready papers due.
September 2026
2 October 2026		 RecSys Challenge Workshop at ACM RecSys 2026.