Algorithmic Collective Action in Recommender Systems: Promoting Songs by Reordering Playlists

We investigate algorithmic collective action in transformer-based recommender systems. Our use case is a music streaming platform where a collective of fans aims to promote the visibility of an underrepresented artist by strategically placing one of their songs in the existing playlists they control. We introduce two easily implementable strategies to select the position at which to insert the song with the goal to boost recommendations at test time. The strategies exploit statistical properties of the learner by targeting discontinuities in the recommendations, and leveraging the long-tail nature of song distributions. We evaluate the efficacy of our strategies using a publicly available recommender system model released by a major music streaming platform. Our findings reveal that through strategic placement even small collectives (controlling less than 0.01\% of the training data) can achieve up to $40\times$ more test time recommendations than an average song with the same number of training set occurrences. Focusing on the externalities of the strategy, we find that the recommendations of other songs are largely preserved, and the newly gained recommendations are distributed across various artists. Together, our findings demonstrate how carefully designed collective action strategies can be effective while not necessarily being adversarial.

Paper

Similar papers

Peer review

Reviewer tKsL5/10 · confidence 4/52024-06-19

Summary

This paper explores the possibility of boosting the recommendation of a song in an automatic playlist continuation system by using a collective strategy of adding the song into the training playlists of the APC at a specific position. The paper shows that adopting a strategy that targets low frequency context makes it possible to very significantly boost the exposition of the song in the output of the APC.

Strengths

- Very interesting idea which is novel in the domain of music recommendation - Quite surprising experimental result that is worth sharing (possibility of boosting exposition of a song with the DirLof strategy).

Weaknesses

- Limited scope: - The idea is tested in the limited context of automatic playlist continuation and is likely not adaptable to broader applications. - Only one APC model was tested, so the results may be specific to this model. It would have been a good idea to test the impact of the collective action on other models (such as the baseline proposed in the reference paper for the APC model). Also, it’s unclear whether the effect would be robust to hyperparameter changes in the APC model. - the method is tested in a static context (not in the real world), so the actual impact of the method on usage (for instance, whether a user exposed through the APC to a song boosted by the collective strategy would listen to it or not) is not tested (this would require access to the production of a streaming music service, though) - Very limited insights on the design of the efficient strategy (DirLof) and why it works. The presentation of the strategy is actually quite unclear, while it’s central to the paper. - There may be other simpler baselines that would be worth testing. The paper shows that inserting the song at the end of the playlists is less performance than inserting the song randomly, which suggests that inserting it earlier in the playlist sequence may help. So a baseline that would always insert the song at the beginning of the sequence could have decent results (the first place is likely to be avoided as it would mean that the song would never appear as the target in the training of the APC, but low positions that appear regularly as targets in the training could be considered).

Questions

- Could the authors rephrase / clarify the definition of the DirLof strategy and rationale behind it? something which is very surprising is that “the collective selects the anchor songs s0 with the smallest overall song frequency”. But it’s likely that smallest song frequency is for song appearing only once among playlists and these song are then mostly noise in playlists and it’s hard to get why those anchors would help improving recommandation of the inserted song. Also, statistics on songs with low frequency are likely much more difficult to obtain (the statistics will be much less reliable than for popular songs). - It’s very unclear what the distributions over playlists are (P, P0, P*). It seems it’s almost used as a synonym of dataset, and the expectations are actually just empirical expectations. I’d like the author to clarify this point and to explain why it’s necessary to introduce such notations. - In table 2, it’s unclear how statistics are estimated when there is no training data to compute them from (while there is still a significant amplification).

Rating

5

Confidence

4

Soundness

3

Presentation

4

Contribution

1

Limitations

- As mentioned in the weaknesses, there are other simple baselines that would be worth testing. That would help support the claim that the specially designed DirLof strategy is actually efficient. - Also the authors should comment on the possible transferability of their results to other APC models, and to modifications of hyperparameters of the model. On the ethical side, the method is a bit too much presented as a way for artists to attack a recommender system, somewhat rerank their songs and make money out of it, which has some important ethical implications in terms of fairness among artists: As the payment system of most streaming services is based on subscription (so a fixed amount of money), artificially boosting an artist comes at the detriment of other artists. I would then rather avoid this message of "opportunity for artists" in the paper (and even in the title) and turn it instead as a warning to music streaming platforms that recommender systems may lack robustness to attacks and that they should be aware of it and take action. I think presenting the paper in this second way would solve this ethical issue, but the current message is, to me, ethically borderline. Minor comments - The authors claim that “most large platforms have shifted from relying on collaborative-based models for APC to building deep learning based recommenders …” but 1) the references are only Spotify based 2) the references are research papers which doesn’t mean that the shift actually happened and that platforms no longer use collaborative-based models part in their complex architecture. - In equation (1): - The authors likely want to specify that the recommended song shouldn’t be part of the playlist i.e S’ **∩** p = **∅** in the argmax - The argmax actually corresponds to the top K elements in terms of similarity, which is quite straightforward. But the sum in the equation makes it not very clear at first read. It could be worth it to state it explicitly before the equation. - There are several notations that use the letter h (song embedding, playlist embedding, playlist mapping). I think using different letters may help make things clearer. - In Figure 3, it seems like s* is subtituted to s_i (there is no s_i in (b)) while just before equation 4, it’s said that only insertions are considered. - “*Existing data poisoning techniques, for example, operate in different settings, not complying with Constraint 1 and typically assuming white-box access to the model or involving test-time profile manipulations.*” ⇒ this needs references. - *“Thus, collective action can make a tremendous difference for these artists: suppose an artist’s song is streamed 10,000 times, yielding a revenue for the artist of $40 for royalties of $0.004 per stream [32]; an amplification of 25 would increase this revenue to $1, 000.”* I understand this is purely illustrative, but the figures won’t reflect much the truth. First 0.004$ per stream is an average that depends on the platform, but also on the kind of registration of the user and on the country of the user with quite important variations. Second, several platforms (Spotify, Deezer…) are shifting to a payment model were all streams don’t have the same value, especially recommended streams are “discounted”, and some “real” artist may get a boost. Once again, I get it’s for illustrative purpose, but it brings confusion on how music payment system works, and then should be avoided in my opinion. - *“Moreover, when considering s* as a relevant recommendation, collective action even enhances the system’s performance”* Isn’t this completely obvious? if presentation of s* is boosted (which happens, given the amplification), other recommendations are barely affected (as shown on figure 6, solid line), and you consider s* a valid recommendation, then for sure recommendation metrics will increase, won’t they?

Reviewer tKsL2024-08-12

Acknowledgement

I'd like to thank the authors for the very detailed answer to my review. Most of my concerns are now addressed. However, I have to disagree on one of the most important ones: presenting the proposed method as a way to empower users to support their artist is, to me, oversimplistic and a bit disconnected from the reality of streaming services. Collective actions (not linked to recommendation) to divert revenue has been widely used in the music streaming industry, first by legit artists (see the Sleepify album by vulfpeck) but then (much more widely) by fraudsters. Both cases are unfair because they trick the payment system, but it's obviously even worse when done by fraudsters. The proposed method also concerns tricking royalties payment through the recommender system, which raises important fairness issues and is obviously also open to fraudsters (who usually control many accounts) and could also be used by major labels (that have strong marketing power) to divert revenue. I think then the work has critical ethical implications that must be discussed and acknowledged and that the positioning of the proposed method as an opportunity is problematic. I won't change my rating if this aspect is not considered in the paper.

Authorsrebuttal2024-08-13

We are glad we could address most of your concerns with the additional experiments and clarifications. Let us follow up on the last one related to the potential for abuse of collective actions strategies. We agree with the reviewer that misuse is a concern and should not be understated. But we still think it is important that the potential for abuse does not prevent the community from working on algorithmic collective action as an interesting and valuable research direction. That being said, we have revisited the paper with your comments in mind, and we can see how it could benefit from additional discussion and a broader perspective. We are happy to use the additional page in the final version to move the broader impact statement to the main body, as suggested by the ethics review, to acknowledge the limitations more prominently. While we have mentioned the threat that a single individual could control many playlists, we will also be more specific that a popular artist could have an advantage to gather large collectives. To further mitigate your concern we will also complement this by expanding more on the study of negative externalities on other artists in Section 4.3. We hope this is satisfactory and you consider voting for acceptance after these changes.

Reviewer tKsL2024-08-13

Thanks for the extra comment. Based on this, I will increase my ratings.

Reviewer fQFx4/10 · confidence 4/52024-07-08

Summary

They proposed a strategy for the streaming platform users to collectively act in order to promote targeted songs. The promotion efficacy is measured by the targeted songs' recommendation frequency boost at testing time. This strategy is approved to be effective through simulation experiment. Another finding is that this strategy has minimum impact to the performance of the recommendation system as a whole, i.e., by preserving user experience to other non-targeted songs.

Strengths

* This is a novel idea, presented in an interesting domain with a good amount of related work. * The writing of the paper is clear. The motivation is sound. * The experiment performed by the author successfully verified the efficacy of the collective strategy.

Weaknesses

* Limited scope. See limitations below. * Lack of technical novelty and contribution. The evaluation result would be a good report but I would recommend the author to seek publication in a different conference.

Questions

The context selection methods discussed in section 3.2 seems pretty arbitrary to me. What reason or motivation does it make you choose the `InClust` or `DirLoF` approach? How would you approve these are guaranteed to work? What if I say a good approach could be to find the most similar song (embedding similar higher than x) that is at least y popular and place our targeted song z ( z could be -1, 1, 2, 3) positions after it?

Rating

4

Confidence

4

Soundness

2

Presentation

3

Contribution

2

Limitations

The strategy could be effective but it is built on top of the assumption that the serving platform has deployed minimum control against collective behavior. As I am aware, there are various user-side anomaly detection mechanisms usually deployed in production in streaming services like what is described in the paper. As an example, there could be real time monitoring of such collective behaviors. Such an anomaly detection system could be constructed by building a user-entity graph. In this use case, we could use artist or songs as the entity. When in a short period of time, there are bursty events that are related to an entity that happens (user promoting a song in their playlist), it could be an important indicator that something unusual happens (because this is not a popular song, it does not receive so many promotes on average). A follow-up action in the system could tag those relevant data as "spurious" thus they never go to the training data of the transformer model (for continuous training). As a result, the collective action could have a significantly less effect when such monitoring is turned on; or depending on what threshold they are set up.

Reviewer 1cCC8/10 · confidence 5/52024-07-08

Summary

The paper shows that strategic collective action by a small fraction of the population can lead to significant amplification of a particular song in a recommender system. The authors propose two strategies for the collective (for a transformer-based song recommender) that achieve this amplification.

Strengths

- The setup is original and focuses on strategic collective action by a fraction of users in a recommender system and how this can increase a target song’s reach. - The performance loss for the platform is negligible, the collective’s strategies are not adversarial (for e.g. fake profiles, artificial perturbations) and based on 1-edit distance to the original playlist. The paper shows that recommendations are largely preserved.

Weaknesses

- The experiments follow the MF-Transformer in [7], to make the paper self-contained it would be beneficial to have a description of $\phi(.)$ and of $g(.)$ and the loss function in Section 2.1 or the Appendix. - I found the strategies in Sec 3.2 hard to parse, perhaps a figure showing the original playlist and the possible changes a user in the collective can do under the two strategies would be helpful. - Minor: I think the notation h(.) is overloaded for the song/playlist embedding in 2.1 and for the strategy mapping which inserts s* into a playlist. Also, fig 6 could use different colors for the different strategies.

Questions

- Can you clarify the intuition and communication required by the collective for the two playlist modification strategies in 3.2? InClust seems to place the target song before the most popular songs in the collective’s playlist whereas DirLoF places the target song after a low popularity song? Both strategies provide amplification in the experiments. How much communication among the collective do both require, is one modification easier practically? - Section 1.1 mentions “Our strategies exploit the properties of the attention function in the transformer model, without requiring knowledge of the model parameters”, can you elaborate?

Rating

8

Confidence

5

Soundness

4

Presentation

3

Contribution

4

Limitations

The authors discuss limitations in Appendix A.

Reviewer 9a2T8/10 · confidence 4/52024-07-13

Summary

This research work proposes a novel solution to promote songs on music streaming platforms strategically. Under the following assumptions: 1. Fans can collaborate to promote a specific song by collectively reordering playlists. 2. The visibility of a song in a playlist affects its recommendation frequency. 3. Users are influenced by the position of songs in playlists when making listening choices. 4. The impact of collective action on song visibility is measurable and significant. The authors suggest that fans strategically reorder playlists to promote a targeted song, thereby increasing its visibility in the recommender system. By leveraging algorithmic collective action, even small groups of fans can substantially impact the recommendation frequency of the promoted song. This strategy aims to enhance the visibility (capability of being discovered) of songs and artists, which will benefit both fans and musicians in the music streaming industry. The evaluation focuses on quantifying the amplification of recommendations achieved by strategically placing songs in playlists, using metrics such as recommendation probability and change in the number of recommendations for a song. The evaluation also includes the impact on the recommendations of other songs and the overall performance of the recommender system. The analysis of results reveals that the collective action strategies can lead to a substantial increase in the recommendation frequency of the targeted song, with up to 25x higher recommendation probability compared to average songs. The main contributions are: 1. The paper introduces **two innovative collective** action strategies where participants strategically insert a target song into their playlists to promote an emerging artist. These strategies aim to increase the recommendations of the targeted song at test time, thereby boosting the artist's visibility on the platform. 2. The research demonstrates that even **small collectives**, controlling less than 0.01% of the training data, **can achieve significant amplification of recommendations** by strategically placing songs in playlists. This finding highlights the effectiveness of algorithmic collective action in promoting songs without major disruptions to the user experience. 3. Preservation of Other Recommendations, as the study reveals that while promoting a specific song through collective action, the recommendations of other songs are largely preserved. This indicates that the proposed strategies can enhance the visibility of targeted songs without significantly compromising the overall recommendation system's performance.

Strengths

- Its innovation is a significant strength as it provides a new approach to increasing the visibility of emerging artists in music streaming platforms. - Empirical Validation and Real-World Application: the research is empirically validated using an open-source APC model deployed on Deezer, a platform with millions of users. - important result: the study demonstrates that even small collectives can achieve substantial amplification of recommendations by strategically placing songs in playlists - The findings show the potential for diverse artist promotion, which can make fairer use of the platforms but also fights against the long-tail problem in recommender systems. It can also help the serendipity effect.

Weaknesses

- The paper assumes that users are influenced by the position of songs in playlists when making listening choices. This assumption **may oversimplify user behavior and overlook other factors** that influence song recommendations and user engagement, potentially leading to biased results.

Questions

Even if the problem is not exactly the same, could you relate your work with the one described in this cite: Walid Bendada, Guillaume Salha-Galvan, Thomas Bouabça, and Tristan Cazenave. 2023. A Scalable Framework for Automatic Playlist Continuation on Music Streaming Services. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '23). Association for Computing Machinery, New York, NY, USA, 464–474. https://doi.org/10.1145/3539618.3591628 ?

Rating

8

Confidence

4

Soundness

3

Presentation

3

Contribution

4

Limitations

The paper does not explicitly discuss possible limitations of the approach to addressing problems of privacy and fairness. However, considering the ethical implications of data manipulation and collective action in recommender systems is crucial for ensuring transparency and equity in algorithmic interventions.

Reviewer 1cCC2024-08-07

Acknowledging and raising score.

Thanks for addressing my questions comprehensively and for the figures for the two strategies, I am raising my score to a strong accept. As a general note directed to another reviewer, it would be nice if as a community we provide constructive feedback and improvement, and not attacks like "good as a report ... seek publication in a different conference".

Authorsrebuttal2024-08-13

Thank you for the assessment. We will use the additional page in the final version to move the broader impact statement to the main text.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC