![\"\"\/](\"https:\/\/research.atspotify.com\/wp-content\/themes\/spotify\/images\/icon.png\"){kind=icon}
\n <\/div>\nApril 08, 2025<\/span> Published by Enrico Palumbo, Gustavo Penha, Andreas Damianou, Jos\u00e9 Luis Redondo Garc\u00eda, Timothy Christopher Heath, Alice Wang, Hugues Bouchard, Mounia Lalmas<\/p>\n<\/p><\/div>\n Imagine asking your music app to \u201cplay some old-school rock ballads to relax<\/em>\u201d and immediately receiving the right observe suggestion. This sort of customized music discovery is the objective of Text2Tracks, a brand new system primarily based on generative AI designed to enhance how music is really helpful primarily based in your selection of phrases.<\/p>\n Some current music suggestion options\u2014reminiscent of these seen in Spotify\u2014have begun to include off-the-shelf Large Language Models (LLMs) to course of pure language prompts.\u00a0 For instance, LLMs can generate artist names and track titles<\/strong> as their output (e.g. User: \u201cCan you recommend some chill songs for me? LLM: \u201cSure, I can recommend While intuitive, the method of figuring out tracks by way of their title has a number of limitations:<\/p>\n To overcome these challenges, we developed Text2Tracks<\/strong>, a mannequin that introduces a extra environment friendly and efficient strategy to advocate music by fine-tuning an LLM to instantly generate optimized observe identifiers (IDs) given a music suggestion immediate.<\/p>\n Figure 1: Text2Tracks supplies music suggestion by producing observe IDs which can be related to the person\u2019s immediate.\u00a0<\/em><\/p>\n \u200b\u200bRather than producing precise track titles and matching them to a database, Text2Tracks makes use of Generative Retrieval<\/strong>: a Generative AI method the place the system is educated to generate observe IDs instantly from textual content prompts. These IDs instantly determine songs within the music catalog and permit for quicker, extra correct suggestions.\u00a0<\/p>\n Here\u2019s the way it works (Fig. 1):<\/p>\n A key distinction between Text2Tracks and off-the-shelf LLMs is that Text2Tracks undergoes an in depth fine-tuning course of tailor-made particularly for music suggestion. We fine-tune the mannequin utilizing a big corpus of chosen playlists, the place lengthy and descriptive playlist titles (e.g., \u201cenergetic rock vibes,\u201d \u201cchill relaxing at the beach\u201d) function a proxy for pure language music prompts. This permits the mannequin to study associations between textual descriptions and related observe alternatives.<\/p>\n To consider the effectiveness of this method, we examine the mannequin\u2019s suggestions to playlists curated by skilled editors, utilizing the hits@10 metric to measure how typically the mannequin\u2019s prime predictions align with professional alternatives. This fine-tuning step helps the mannequin higher seize the nuances of music-related language and generate extra related and contextually acceptable observe suggestions.<\/p>\n One of probably the most progressive features of Text2Tracks is the way it assigns IDs to songs. We experimented with three completely different methods to seek out the easiest way to symbolize tracks (Tab. 1):<\/p>\n We discovered that Semantic IDs<\/strong> constructed on prime of collaborative filtering vectors<\/strong> labored the very best. This highlights the flexibility of Semantic IDs and of the underlying vector area to seize extra nuances in observe representations, leading to a better accuracy within the remaining set of suggestions.\u00a0 The Artist Track Integer<\/em> technique got here in second, which is outstanding contemplating it\u2019s comparatively easy to implement\u2014it solely makes use of the artist and observe and doesn\u2019t want an embedding area. This reveals how vital the artist-track hierarchy is for modeling observe identifiers and the way a lot it could possibly assist the mannequin study to make good observe suggestions.\u00a0<\/p>\n It can also be price noting that the Artist Track Integer<\/em> outperforms the Artist Track Name<\/em> technique, which is usually utilized in many eventualities, reminiscent of off-the-shelf LLM suggestions. This <\/em>additional underscores the benefit of constructing structured observe identifiers reasonably than figuring out songs by way of their titles.<\/p>\n Tab. 1: hits@10 for various methods to create observe identifiers within the generative retrieval setting<\/em><\/p>\n We in contrast Text2Tracks to different textual suggestion strategies, together with:<\/p>\n Tab. 2: hits@10 for Text2Tracks in comparison with well-known textual content retrieval approaches We see that Text2Tracks considerably outperformed these baselines<\/strong>, reaching 127%<\/strong> higher accuracy than the closest competitor. When wanting into why<\/strong> it really works so nicely, we qualitatively observe that it tends to advocate extremely related and canonical tracks for broad prompts, matching the person\u2019s expectation. For occasion, when requested for \u201cChristmas classics<\/em>\u201d, Text2Tracks reliably retrieved probably the most iconic vacation tracks, whereas different programs supplied extra generic vacation songs.<\/p>\n In this submit, we launched Text2Tracks, a brand new method to music suggestion that generates observe IDs instantly from a person immediate, delivering quicker and extra correct outcomes. Trained on playlist information and leveraging Semantic IDs constructed on prime of collaborative filtering embeddings, Text2Tracks outperforms present programs for prompt-based music suggestion.<\/p>\n These outcomes spotlight the potential of generative fashions to complement music search and suggestion on Spotify. We are excited to proceed advancing these applied sciences, with the objective of constructing a unified generative AI system that enhances a number of aspects of music suggestion. Ultimately, we goal to make music discovery extra customized, intuitive, and pleasing for everybody.<\/p>\n![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2025\/04\/RS069-Text2Tracks_-Improving-Prompt-based-Music-Recommendations-with-Generative-Retrieval-no-logo.png\"){kind=logo}
Why Current Methods Fall Short<\/strong><\/h2>\n
\n
![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2025\/04\/image1-700x410.png\")
<\/figure>\n<\/div>\nHow Text2Tracks Works<\/strong><\/h2>\n
\n
How Text2Tracks is Trained and Tested<\/strong><\/h2>\n
The Role of Track IDs<\/strong><\/h2>\n
\n
![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2025\/04\/Screenshot-2025-04-07-at-11.28.53-700x218.png\")
<\/figure>\nHow Text2Tracks Compare to Other Systems<\/strong><\/h2>\n
\n
![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2025\/04\/Screenshot-2025-04-07-at-11.29.00-700x304.png\")
<\/figure>\n<\/div>\n
<\/em><\/p>\nLooking forward<\/strong><\/h2>\n