Within Attention
Do Music Algorithms Broaden Taste?
Recommendation systems can help listeners explore, but optimisation may still favour music that already has strong attention signals.
On this page
- How recommendations help listeners navigate abundance
- Why popularity bias matters for the long tail
- The trade off between satisfaction, diversity, and fairness
Page outline Jump by section
Introduction
Recommendation algorithms can widen music taste, but they do not do so automatically. Their core purpose is usually to help listeners navigate overwhelming abundance by predicting what they are likely to enjoy next. In a catalogue containing tens or hundreds of millions of tracks, that function is genuinely useful. Many listeners discover artists, genres and scenes they would never have found through manual searching alone. Yet the same systems often rely on signals such as previous listening behaviour, engagement rates and existing popularity. As a result, they can expand taste in some directions while narrowing it in others. The central fairness question is not whether algorithms can recommend unfamiliar music, but whether they can do so without systematically favouring music that already attracts attention. Research suggests that achieving both satisfaction and fair exposure remains one of the hardest design problems in music recommendation. [PMC]pmc.ncbi.nlm.nih.govPMCFairness in Music Recommender Systems: A Stakeholder…by K Dinnissen · 2022 · Cited by 54 — This mini review, therefore, outlines cu…
How Recommendations Help Listeners Navigate Abundance
Music recommendation systems emerged because modern catalogues are too large for human navigation alone. Most systems combine signals from a listener’s own history with patterns observed across millions of other users. If people with similar listening habits enjoy a particular artist or track, the system may recommend it. Content-based approaches also analyse musical characteristics, allowing recommendations that sound similar even when listeners have never encountered the artist before. [Wikipedia]WikipediaRecommender systemRecommender system
In principle, this creates opportunities for discovery. A listener who enjoys one jazz pianist might be introduced to contemporary players from another country. Someone interested in indie rock could encounter adjacent genres such as dream pop or post-punk. Recommendation systems can lower the cost of exploration by reducing the effort required to search vast catalogues manually. Fairness researchers note that recommendation systems can benefit multiple stakeholders simultaneously: listeners gain access to relevant music, while artists gain potential exposure beyond their existing audience. [PMC]pmc.ncbi.nlm.nih.govResearch has shown that recommender systems are typically biased towards popular items, which leads to less popular items being underrepr…
Importantly, diversity is not merely a cultural ideal. Research conducted at Spotify found that listeners with broader consumption patterns tend to exhibit stronger long-term engagement with the platform. This suggests that helping users explore beyond narrow habits may improve both user experience and platform outcomes. [Spotify Research]research.atspotify.comalgorithmic effects on the diversity of consumption on spotifySpotify ResearchAlgorithmic Effects on the Diversity of Consumption on Spotify3 Dec 2020 — In this work, we analyze our users through the…
Why Popularity Bias Matters for the Long Tail
The difficulty is that recommendation systems often learn from attention signals that are already unevenly distributed. Popular songs generate more plays, saves, shares and behavioural data. Because algorithms are trained on such signals, they frequently conclude that already-popular music is the safest recommendation.
Researchers describe this pattern as popularity bias. Studies in music recommendation repeatedly find that popular artists and tracks receive disproportionate recommendation exposure, while less-known music is underrepresented. This can create a feedback loop: music that gains attention becomes easier for algorithms to recommend, which generates more attention, which further strengthens its recommendation signals. [PMC]pmc.ncbi.nlm.nih.govPMCFairness in Music Recommender Systems: A Stakeholder…by K Dinnissen · 2022 · Cited by 54 — This mini review, therefore, outlines cu… [arXiv]arxiv.orgOpen source on arxiv.org.
This matters because streaming catalogues contain enormous long tails of niche and emerging music. Availability alone does not guarantee visibility. If recommendation engines primarily reinforce existing popularity, they may preserve the appearance of infinite choice while directing listening toward a relatively small portion of the catalogue. Research on recommender systems more broadly warns that popularity bias can produce cumulative reinforcement effects over time, concentrating attention instead of distributing it. [Springer Link]link.springer.comSpringer LinkA survey on popularity bias in recommender systemsby A Klimashevskaia · 2024 · Cited by 191 — In this paper, we discuss the…
The fairness issue affects listeners as well as artists. One study of music recommendation found that users whose tastes diverged from mainstream preferences tended to receive poorer recommendations than users whose tastes aligned with already-popular music. In other words, the same bias that disadvantages niche artists can also disadvantage listeners seeking less conventional music. [arXiv]arxiv.orgSource details in endnotes.
Why More Recommendations Do Not Always Mean Broader Taste
A common assumption is that recommendation systems naturally increase diversity because they expose users to more music. The evidence is more complicated.
Spotify researchers studying large-scale listening behaviour found that algorithmically driven listening was associated with reduced diversity of consumption, even though diverse listening itself correlated with positive user outcomes. The finding does not mean algorithms always narrow taste, but it demonstrates that recommendation-driven listening can become concentrated around familiar patterns. [Spotify Research]research.atspotify.comalgorithmic effects on the diversity of consumption on spotifySpotify ResearchAlgorithmic Effects on the Diversity of Consumption on Spotify3 Dec 2020 — In this work, we analyze our users through the…
This happens because recommendation systems are usually optimised around prediction accuracy and engagement. If a listener repeatedly plays a particular style, recommending similar tracks often produces reliable results. The algorithm learns what is safe rather than what is surprising. Over time, listeners may encounter a stream of music that varies within a comfort zone but rarely challenges it.
Critics of streaming culture argue that this dynamic can create a subtle form of cultural narrowing. The issue is not censorship or exclusion; rather, the path of least resistance tends to lead toward music that resembles what listeners already know. The resulting experience can feel highly personalised while still limiting the range of exploration. [GOV.UK]GOV.UKThe impact of algorithmically driven recommendation…by D Hesmondhalgh · Cited by 56 — The impact of streaming platforms on musical pro…
The Trade-Off Between Satisfaction, Diversity, and Fairness
The central challenge is that recommendation systems must balance competing goals.
- Satisfaction: Deliver music listeners are likely to enjoy immediately.
- Diversity: Introduce enough novelty to prevent stagnation.
- Fairness: Ensure lesser-known artists and genres receive meaningful opportunities for exposure.
These goals often conflict. Recommending only familiar favourites may maximise short-term engagement but reduce diversity. Recommending too much unfamiliar music may increase diversity while lowering immediate satisfaction. Promoting underexposed artists may improve fairness but can reduce predictive accuracy if the system has less behavioural data about them. [Music Information Retrieval Transactions]transactions.ismir.netMusic Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv…
Research consistently finds that the most accurate recommendation models are often among the most popularity-biased. This creates a practical dilemma for platforms whose business models depend heavily on user engagement metrics. [arXiv]arxiv.orgSource details in endnotes.
Fairness researchers increasingly argue that recommendation quality should not be judged solely by accuracy. Diversity, serendipity, artist exposure and user control are now treated as legitimate design objectives alongside engagement metrics. [Music Information Retrieval Transactions]transactions.ismir.netMusic Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv…
What Fairer Music Recommendation Might Look Like
Fairness in music recommendation does not necessarily require equal exposure for every track. Instead, it involves designing systems that avoid systematically disadvantaging particular artists, genres or listener groups.
Several approaches have emerged:
- Introducing diversity constraints that deliberately mix familiar and unfamiliar music.
- Limiting the dominance of popularity signals in ranking models.
- Providing listeners with controls that let them choose how exploratory recommendations should be.
- Measuring success using fairness and diversity metrics rather than engagement alone.
- Considering artist exposure as a design goal alongside listener satisfaction. [Music Information Retrieval Transactions]transactions.ismir.netMusic Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv… [PMC]pmc.ncbi.nlm.nih.govResearch has shown that recommender systems are typically biased towards popular items, which leads to less popular items being underrepr…
Recent experimental work suggests that giving users direct control over factors such as popularity, genre diversity and artist representation can improve perceived fairness and user satisfaction simultaneously. Rather than forcing a single definition of fairness, such systems allow listeners to decide how adventurous or balanced they want recommendations to be. [Utrecht University]research-portal.uu.nlUtrecht UniversityUser-Driven Fairness in Music Recommendations: Effects…by SN Khan · 2025 — This study investigates how user-driven c…
Fairness Is a Design Choice, Not a Technical Guarantee
Recommendation algorithms are capable of broadening musical horizons. They can connect listeners with obscure artists, revive forgotten catalogues and expose audiences to scenes that would otherwise remain invisible. Yet the same systems can also reinforce existing attention patterns because popularity itself is a powerful source of data.
The key lesson is that discovery does not emerge automatically from abundance. Whether recommendation systems widen music taste fairly depends on the objectives they optimise for. If success is measured only by immediate engagement, popularity bias is difficult to avoid. If diversity, artist exposure and listener exploration are treated as explicit goals, algorithms can become tools for cultural discovery rather than merely engines of attention concentration. [GOV.UK]GOV.UKThe impact of algorithmically driven recommendation…by D Hesmondhalgh · Cited by 56 — The impact of streaming platforms on musical pro… [3PMC 3Music Information]transactions.ismir.netMusic Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv… Retrieval Transactions](#endnote-15 “Snippet: Music Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv”)
Amazon book picks
Further Reading
Books and field guides related to Do Music Algorithms Broaden Taste?. Use these as the next step if you want deeper reading beyond the article.
Weapons of Math Destruction
Provides an accessible framework for understanding fairness concerns in recommendation systems.
The Filter Bubble
First published 2011. Subjects: World Wide Web, Invisible Web, Information organization, Social aspects, Subject access.
Algorithms of Oppression
First published 2018. Subjects: Google, Sociological aspects, Search engines, Discrimination, COMPUTERS / Web.
Mood Machine
Directly addresses recommendation systems, playlists, and platform incentives.
eBay marketplace picks
Marketplace Samples
Example marketplace items related to this page. Use the search link to explore similar finds on eBay.
Endnotes
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC9353048/Source snippet
PMCFairness in Music Recommender Systems: A Stakeholder...by K Dinnissen · 2022 · Cited by 54 — This mini review, therefore, outlines cu...
-
Source: Wikipedia
Title: Recommender system
Link: https://en.wikipedia.org/wiki/Recommender_system -
Source: GOV.UK
Link: https://www.gov.uk/government/publications/research-into-the-impact-of-streaming-services-algorithms-on-music-consumption/the-impact-of-algorithmically-driven-recommendation-systems-on-music-consumption-and-production-a-literature-reviewSource snippet
The impact of algorithmically driven recommendation...by D Hesmondhalgh · Cited by 56 — The impact of streaming platforms on musical pro...
-
Source: research.atspotify.com
Title: algorithmic effects on the diversity of consumption on spotify
Link: https://research.atspotify.com/algorithmic-effects-on-the-diversity-of-consumption-on-spotifySource snippet
Spotify ResearchAlgorithmic Effects on the Diversity of Consumption on Spotify3 Dec 2020 — In this work, we analyze our users through the...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC7148048/Source snippet
Research has shown that recommender systems are typically biased towards popular items, which leads to less popular items being underrepr...
-
Source: arxiv.org
Link: https://arxiv.org/abs/1912.04696 -
Source: arxiv.org
Title: arXiv Unfair Exposure of Artists in Music Recommendation
Link: https://arxiv.org/abs/2003.11634 -
Source: link.springer.com
Link: https://link.springer.com/article/10.1007/s11257-024-09406-0Source snippet
Springer LinkA survey on popularity bias in recommender systemsby A Klimashevskaia · 2024 · Cited by 191 — In this paper, we discuss the...
-
Source: arxiv.org
Title: arXiv A Survey on Popularity Bias in Recommender Systems
Link: https://arxiv.org/abs/2308.01118 -
Source: arxiv.org
Link: https://arxiv.org/abs/2208.09517Source snippet
arXivExploring Popularity Bias in Music Recommendation Models and Commercial Steaming ServicesAugust 19, 2022...
Published: August 19, 2022
-
Source: arxiv.org
Link: https://arxiv.org/pdf/1912.04696Source snippet
Research has shown that recommender systems are typically biased towards popular items, which leads to less popular items being un...Rea...
-
Source: arxiv.org
Link: https://arxiv.org/html/2504.04752v1Source snippet
Investigating Popularity Bias Amplification in...7 Apr 2025 — This work summarizes our research on investigating the amplification of po...
-
Source: arxiv.org
Link: https://arxiv.org/pdf/2308.14601Source snippet
Mitigating Popularity Bias For Music Discoveryby R Salganik · 2023 · Cited by 3 — To mitigate this issue we propose a domain-aware, indiv...
-
Source: arxiv.org
Link: https://arxiv.org/pdf/2308.01118Source snippet
Recommender systems help people find relevant content in a personalized way. One main promise of such systems is that they are.Read more...
-
Source: transactions.ismir.net
Link: https://transactions.ismir.net/articles/10.5334/tismir.106Source snippet
Music Information Retrieval TransactionsDiversity by Design in Music Recommender Systemsby L Porcaro · 2021 · Cited by 33 — In this overv...
-
Source: research.atspotify.com
Title: algorithmic effects on the diversity of consumption on spotify
Link: https://research.atspotify.com/publications/algorithmic-effects-on-the-diversity-of-consumption-on-spotifySource snippet
Spotify ResearchAlgorithmic Effects on the Diversity of Consumption on Spotify1 Apr 2020 — However, we also find that algorithmically-dri...
-
Source: research-portal.uu.nl
Link: https://research-portal.uu.nl/en/publications/user-driven-fairness-in-music-recommendations-effects-on-experien/Source snippet
Utrecht UniversityUser-Driven Fairness in [Music Recommendations]({{ 'algorithms-a7ef46/' | relative_url }}): Effects...by SN Khan · 2025 — This study investigates how user-driven c...
-
Source: research.atspotify.com
Link: https://research.atspotify.com/publications?category=evaluationSource snippet
| Spotify ResearchThe Engagement-Diversity Connection: Evidence from a Field Experiment on Spotify. David Holtz, Benjamin Carterette, Pra...
Additional References
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/340573520_The_Unfairness_of_Popularity_Bias_in_Music_Recommendation_A_Reproducibility_StudySource snippet
The Unfairness of Popularity Bias in Music Recommendation15 Apr 2020 — Research has shown that recommender systems are typically biased t...
-
Source: hcai.at
Link: https://hcai.at/publications/2021_RecSysLBR_PopBiasGender/Source snippet
Analyzing Item Popularity Bias of Music Recommender SystemsWe focus on music recommendation and conduct experiments on the recently relea...
-
Source: medium.com
Link: https://medium.com/%40kamalmeet/popularity-bias-in-recommendation-engines-2542d1cdb353Source snippet
Popularity Bias in Recommendation EnginesPopularity bias refers to a recommender system's tendency to over-recommend items that are alrea...
-
Source: blogs.biomedcentral.com
Link: https://blogs.biomedcentral.com/on-physicalsciences/2021/04/13/algorithm-generated-music-recommendations-low-accuracy-for-fans-of-beyond-mainstream-music/Source snippet
Accuracy for Fans of Beyond-Mainstream Music13 Apr 2021 — However, it is a widely-known problem that recommender systems are prone to pop...
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/384680227_Bypassing_the_Popularity_Bias_Repurposing_Models_for_Better_Long-Tail_RecommendationSource snippet
(PDF) Bypassing the Popularity Bias: Repurposing Models...17 Sept 2024 — We propose a novel approach of repurposing existing components...
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/384754871_Fairness_and_Transparency_in_Music_Recommender_Systems_Improvements_for_ArtistsSource snippet
Fairness and Transparency in Music Recommender SystemsFairMatch: A Graph-based Approach for Improving Aggregate Diversity in Recommender...
-
Source: researchgate.net
Title: 388827947 Effects of algorithmic curation in users’ music taste on Spotify
Link: https://www.researchgate.net/publication/388827947_Effects_of_algorithmic_curation_in_users%27_music_taste_on_SpotifySource snippet
Effects of algorithmic curation in users' music taste on Spotify8 May 2026 — A comprehensive review of the literature reveals that the pr...
Published: May 2026
-
Source: researchgate.net
Title: 341126150 Algorithmic Effects on the Diversity of Consumption on Spotify
Link: https://www.researchgate.net/publication/341126150_Algorithmic_Effects_on_the_Diversity_of_Consumption_on_SpotifySource snippet
Algorithmic Effects on the Diversity of Consumption on Spotify28 Feb 2026 — [3] found that following Spotify's recommendations reduced th...
-
Source: fairmuse.eu
Link: https://fairmuse.eu/wp-content/uploads/2023/12/Tuning-In-A-Comprehensive-Analysis-of-Music-Recommender-Systems-Playlists-and-Algorithmic-Fairness.pdfSource snippet
of being unfair because they sustain or amplify biases and imbalances against some categories of...Read more...
-
Source: researchgate.net
Title: The Diversity of Music Recommender Systems Similarly, Anderson et al
Link: https://www.researchgate.net/publication/359434519_The_Diversity_of_Music_Recommender_SystemsSource snippet
[3] found that following Spotify's recommendations reduced the diversity of the users' listening lists (see also [8]) and that users...
Topic Tree