Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Personalized Audiobook Recommendations at Spotify Through Graph Neural Networks (2403.05185v1)

Published 8 Mar 2024 in cs.IR and cs.LG

Abstract: In the ever-evolving digital audio landscape, Spotify, well-known for its music and talk content, has recently introduced audiobooks to its vast user base. While promising, this move presents significant challenges for personalized recommendations. Unlike music and podcasts, audiobooks, initially available for a fee, cannot be easily skimmed before purchase, posing higher stakes for the relevance of recommendations. Furthermore, introducing a new content type into an existing platform confronts extreme data sparsity, as most users are unfamiliar with this new content type. Lastly, recommending content to millions of users requires the model to react fast and be scalable. To address these challenges, we leverage podcast and music user preferences and introduce 2T-HGNN, a scalable recommendation system comprising Heterogeneous Graph Neural Networks (HGNNs) and a Two Tower (2T) model. This novel approach uncovers nuanced item relationships while ensuring low latency and complexity. We decouple users from the HGNN graph and propose an innovative multi-link neighbor sampler. These choices, together with the 2T component, significantly reduce the complexity of the HGNN model. Empirical evaluations involving millions of users show significant improvement in the quality of personalized recommendations, resulting in a +46% increase in new audiobooks start rate and a +23% boost in streaming rates. Intriguingly, our model's impact extends beyond audiobooks, benefiting established products like podcasts.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (49)
  1. Inductive representation learning in large attributed graphs. arXiv preprint arXiv:1710.09471, 2017.
  2. E. Bernhardsson. Annoy. https://github.com/spotify/annoy.
  3. Translating embeddings for modeling multi-relational data. Advances in neural information processing systems, 26, 2013.
  4. Graph heterogeneous multi-relational recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 3958–3966, 2021.
  5. J. Chicaiza and P. Valdiviezo-Diaz. A comprehensive survey of knowledge graph-based recommender systems: Technologies, development, and contributions. Information, 12(6):232, 2021.
  6. Performance of recommender algorithms on top-n recommendation tasks. In Proceedings of the fourth ACM conference on Recommender systems, pages 39–46, 2010.
  7. Episodes discovery recommendation with multi-source augmentations. arXiv preprint arXiv:2301.01737, 2023.
  8. M. Fey and J. E. Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019.
  9. A survey on knowledge graph-based recommender systems. IEEE Transactions on Knowledge and Data Engineering, 34(8):3549–3568, 2020.
  10. Linkless link prediction via relational distillation. In International Conference on Machine Learning, pages 12012–12033. PMLR, 2023.
  11. Multibisage: A web-scale recommendation system using multiple bipartite graphs at pinterest. arXiv preprint arXiv:2205.10666, 2022.
  12. Inductive representation learning on large graphs. Advances in neural information processing systems, 30, 2017.
  13. Learning both weights and connections for efficient neural network. Advances in neural information processing systems, 28, 2015.
  14. I. Have and B. S. Pedersen. Reading audiobooks. Beyond Media Borders, Volume 1: Intermedial Relations among Multimodal Media, pages 197–216, 2021.
  15. Uber-gnn: A user-based embeddings recommendation based on graph neural networks. arXiv preprint arXiv:2008.02546, 2020.
  16. Redundancy-free computation for graph neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 997–1005, 2020.
  17. M. Kabiljo and A. Ilic. Recommending items to more than a billion people. Retrieved May, 2:2018, 2015.
  18. D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
  19. T. N. Kipf and M. Welling. Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907, 2016.
  20. Grouplens: Applying collaborative filtering to usenet news. Communications of the ACM, 40(3):77–87, 1997.
  21. J. E. Moyer. Audiobooks and e-books: A literature review. Reference and User Services Quarterly, 51(4):340–354, 2012.
  22. Deepwalk: Online learning of social representations. In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 701–710, 2014.
  23. N. Reimers and I. Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 11 2019.
  24. S. Rendle. Factorization machines. In 2010 IEEE International conference on data mining, pages 995–1000. IEEE, 2010.
  25. Deep feature learning for graphs. arXiv preprint arXiv:1704.08829, 2017.
  26. Graph neural networks for friend ranking in large-scale social platforms. In Proceedings of the Web Conference 2021, pages 2535–2546, 2021.
  27. Item-based collaborative filtering recommendation algorithms. In Proceedings of the 10th international conference on World Wide Web, pages 285–295, 2001.
  28. Recommender systems handbook. 2022.
  29. Link prediction with non-contrastive learning. arXiv preprint arXiv:2211.14394, 2022.
  30. Spotify. With audiobooks launching in the u.s. today, spotify is the home for all the audio you love. https://newsroom.spotify.com/2022-09-20/with-audiobooks-launching-in-the-u-s-today-spotify-is-the-home-for-all-the-audio-you-love/, 2022.
  31. Spotify. Spotify premium will include instant access to 150,000+ audiobooks. https://newsroom.spotify.com/2023-10-03/audiobooks-included-in-spotify-premium/, 2023.
  32. Graph attention networks. arXiv preprint arXiv:1710.10903, 2017.
  33. Recommending related products using graph neural networks in directed graphs. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 541–557. Springer, 2022.
  34. A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems, 32(1):4–24, 2020.
  35. Graphsail: Graph structure aware incremental learning for recommender systems. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, pages 2861–2868, 2020.
  36. Tinygnn: Learning efficient graph neural networks. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 1848–1856, 2020.
  37. Mixed negative sampling for learning two-tower neural networks in recommendations. In Companion Proceedings of the Web Conference 2020, pages 441–447, 2020.
  38. Self-supervised learning for large-scale item recommendations. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 4321–4330, 2021.
  39. Sampling-bias-corrected neural modeling for large corpus item recommendations. In Proceedings of the 13th ACM Conference on Recommender Systems, pages 269–277, 2019.
  40. Graph convolutional neural networks for web-scale recommender systems. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pages 974–983, 2018.
  41. Graphsaint: Graph sampling based inductive learning method. arXiv preprint arXiv:1907.04931, 2019.
  42. Heterogeneous graph neural network. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 793–803, 2019.
  43. Agl: a scalable system for industrial-purpose graph machine learning. arXiv preprint arXiv:2003.02454, 2020.
  44. M. Zhang and Y. Chen. Link prediction based on graph neural networks. Advances in neural information processing systems, 31, 2018.
  45. Graph-less neural networks: Teaching old mlps new tricks via distillation. arXiv preprint arXiv:2110.08727, 2021.
  46. Deep learning on graphs: A survey. IEEE Transactions on Knowledge and Data Engineering, 34(1):249–270, 2020.
  47. Learned low precision graph neural networks. arXiv preprint arXiv:2009.09232, 2020.
  48. Graph neural networks: A review of methods and applications. AI open, 1:57–81, 2020.
  49. A fast parallel sgd for matrix factorization in shared memory systems. In Proceedings of the 7th ACM conference on Recommender systems, pages 249–256, 2013.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (14)
  1. Marco De Nadai (26 papers)
  2. Francesco Fabbri (22 papers)
  3. Paul Gigioli (2 papers)
  4. Alice Wang (9 papers)
  5. Ang Li (472 papers)
  6. Fabrizio Silvestri (75 papers)
  7. Laura Kim (5 papers)
  8. Shawn Lin (1 paper)
  9. Vladan Radosavljevic (14 papers)
  10. Sandeep Ghael (1 paper)
  11. David Nyhan (1 paper)
  12. Hugues Bouchard (6 papers)
  13. Mounia Lalmas-Roelleke (2 papers)
  14. Andreas Damianou (28 papers)
Citations (6)
View on Semantic Scholar

Summary

Unveiling Spotify's Approach to Personalized Audiobook Recommendations with Graph Neural Networks

Introduction

Spotify, initially celebrated for its comprehensive music and podcast offerings, has recently broadened its horizon by introducing audiobooks to its platform. This addition, though promising, introduces several challenges in personalization and recommendation. To address these challenges, a novel approach leveraging Graph Neural Networks (GNNs) and a Two-Tower model has been developed, demonstrating a significant improvement in the recommendation quality for audiobooks. This post explores the architecture, findings, and implications of this innovative recommendation system.

Challenges in Audiobook Recommendations

The introduction of audiobooks to an established platform like Spotify entails unique challenges:

  • Data Sparsity: The novelty of audiobooks on Spotify means limited user interaction data, making it harder to generate relevant recommendations.
  • High Stakes for Recommendations: Given that audiobooks initially required purchase before listening, there's a greater emphasis on the accuracy and relevance of recommendations.
  • Need for Scalability: The system must efficiently cater to Spotify’s vast user base without compromising on latency.

Solution Overview: 2T-HGNN

To navigate these challenges, the 2T-HGNN (Two-Tower Heterogeneous Graph Neural Network) model was introduced. The model creatively utilizes existing user preferences for podcasts and music to enhance audiobook recommendations. It encapsulates a dual approach by employing:

  • A Heterogeneous Graph Neural Network (HGNN) that learns nuanced item relationships from user interactions with music and podcasts.
  • A Two Tower (2T) model that ensures scalable and efficient recommendations by generating user and item embeddings, benefitting from low latency and reduced complexity.

The crux of this system lies in its ingenious architecture that separate content relationships and user-item interactions, thereby managing to reduce complexity and enhance scalability.

Empirical Evaluations

The system was rigorously evaluated, depicting a remarkable +46% increase in new audiobook start rates and a +23% boost in streaming rates among Spotify users. These figures not only validate the effectiveness of the 2T-HGNN model but also highlight the potential of leveraging cross-content type interactions (e.g., between podcasts and audiobooks) for recommendations.

Implications and Future Directions

This research holds profound implications for the future of AI in digital content recommendation. The success of the 2T-HGNN model underscores the significance of:

  • Cross-Content Learning: Utilizing user preferences across different content types can significantly improve recommendation systems.
  • Scalable AI Models: The necessity for AI models that are not only accurate but also scalable and efficient in handling vast amounts of data and users.
  • Graph Neural Networks: The potential of GNNs, especially HGNNs, in unraveling complex item relationships and enhancing recommendation quality.

Moreover, the architectural novelty of decoupling content relationships from user-item interactions offers a blueprint for future recommendation systems across various digital platforms.

Concluding Thoughts

The introduction and success of the 2T-HGNN model mark a significant leap toward personalized content recommendations on digital platforms. By leveraging the connections between different types of content and addressing the unique challenges of audiobook recommendations, Spotify has set a new benchmark for personalization in the streaming industry. The insights gained from this research pave the way for future advancements in recommendation systems, promising enhanced user experiences across a myriad of digital services.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Youtube Logo Streamline Icon: https://streamlinehq.com

YouTube