Opening the Black Box: Interpretable Remedies for Popularity Bias in Recommender Systems

Masoud Mansoury; Parviz Ahmadov

REVIEW 4 major objections 7 minor 35 references

Opening the Black Box: Interpretable Remedies for Popularity Bias in Recommender Systems

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read PopSteer shows that popularity bias in a sequential recommender can be attributed to specific sparse-autoencoder neurons and corrected by steering their activations.

desk verdict PopSteer is a clever, effective post-hoc steering method for popularity bias, but the neuron-level interpretability claim rests on synthetic probes that don't isolate popularity from content correlations. read the letter →

arxiv 2508.17297 v1 pith:DO6PVW24 submitted 2025-08-24 cs.IR

Parviz Ahmadov , Masoud Mansoury This is my paper · ORCID

classification cs.IR

keywords popularitybiasrecommendersystemssparseautoencoderneuronsteeringinterpretabilityfairnesssequentialrecommendation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

PopSteer is a post-hoc method for opening up a deep recommender's popularity bias: it trains a sparse autoencoder (SAE) on the embeddings of a pretrained sequential recommender, then uses synthetic user profiles built from popular and unpopular items to find which SAE neurons respond to popularity. The paper claims that suppressing neurons that fire for popular items and amplifying neurons that fire for unpopular items improves exposure fairness, measured by long-tail coverage and the Gini index, while keeping nDCG@10 close to the original model. The point matters because existing debiasing methods reweight scores or rerank lists without explaining what inside the model drives the bias. PopSteer offers both a diagnosis, which neurons encode popularity, and a fine-grained control knob for the fairness-accuracy trade-off.

What carries the argument

The central object is a sparse autoencoder attached to the final layer of a pretrained sequential recommender. It reconstructs each user embedding $\mathbf{x}$ as $\hat{x} = W_{\mathrm{dec}} a + b_{\mathrm{pre}}$ with a hidden activation vector $a$ that keeps only its top-$K$ largest entries, so each active neuron is forced to specialize. The machinery has three moving parts: the synthetic popular/unpopular user profiles used to probe the SAE; a per-neuron standardized effect size $d_j$ computed from the activation difference between the two profile types; and the steering rule $a'_j = a_j \pm w_j \sigma_j$, where weights $w_j$ are normalized magnitudes of $d_j$ and the sign depends on whether the neuron is popularity-aligned or unpopularity-aligned. Together these parts turn the black-box embedding into a list of identifiable, adjustable popularity signals.

What would settle it

Generate synthetic profiles in which popular and unpopular items are matched on genre, category, and release-era distributions before computing the per-neuron effect sizes; if the neurons identified as popularity-positive stop predicting the Gini-index improvement from steering, the attribution is driven by those confounds and not by popularity alone.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that the popularity bias of SASRec can be localized to a small set of interpretable neurons in an SAE trained to reconstruct the model's final-layer user embeddings. After replacing each user's items with random popular items or random unpopular items and recording hidden-layer activations, the authors compute a standardized effect size per neuron; large positive values mean the neuron fires more for popular profiles, large negative values mean it fires more for unpopular ones. PopSteer then adjusts those activations by a normalized, magnitude-weighted shift, boosting $d<0$ neurons and suppressing $d>0$ neurons. On ML-1M and Last.fm, this produces a better fairness-accuracy frontier than reweighting and reranking baselines, and the ablation shows that the same accuracy budget spent on random Gaussian noise does not produce the fairness gains. A direct neuron-deactivation experiment confirms the causal direction: switching off popularity-positive neurons lowers the Gini index, while switching off unpopularity-positive neurons raises it.

Load-bearing premise

The load-bearing premise is that replacing every item in a user's history with a random popular or unpopular item isolates the popularity signal while leaving everything else about the user unchanged; if popularity correlates with item content, category, or timing, the selected neurons may be responding to those confounds rather than to popularity itself.

Editorial extensions

If this is right

PopSteer can be applied after training, without retraining or changing the base recommender, so it fits into deployed systems as a post-hoc fairness layer.
The SAE's per-neuron popularity scores give practitioners an audit trail: a system maintainer can see which internal features carry the bias and how strongly.
Tuning the steering strength $\alpha$ and the number of adjusted neurons $N$ traces a fairness-accuracy frontier, allowing operators to choose how much accuracy they trade for exposure fairness.
The ablation implies targeted steering is doing the work: random perturbation of the same number of neurons at the same nDCG budget does not improve fairness.
The neuron-deactivation experiment turns interpretability into a causal check: the Gini index moves in the expected direction when identified neurons are switched off, confirming the identified neurons are causally linked to popularity bias.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same probe-and-steer recipe should transfer to other biases that can be expressed as synthetic user profiles, such as recency, category, or provider exposure, by swapping the popular and unpopular item sets; the paper only demonstrates popularity.
If popular and unpopular items differ systematically in content features such as genre, language, or release period, the synthetic replacement may let those confounds leak into the effect-size scores; a natural robustness check is to repeat the probe with popular and unpopular items matched on item metadata.
Because the SAE is trained on final-layer embeddings rather than on raw item text, the neurons it labels as popularity-related may actually encode whatever correlates with popularity in the embedding space; comparing steering outcomes across SAE seeds or sparsity levels would test how stable the attribution is.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This paper transfers SAE feature steering from LLMs to sequential recommenders: it trains a sparse autoencoder on SASRec's user embeddings, builds synthetic popular/unpopular profiles to pick neurons via Cohen's d, and then adjusts activations to reduce popularity bias. On ML-1M and Last.fm the method improves long-tail coverage and Gini while keeping nDCG@10 close to the unsteered model, and the random-noise ablation makes a solid case that the targeted selection is what drives the fairness gains. The effectiveness result looks real.

What's genuinely new is the packaging: SAE-based interpretability plus a post-hoc steering knob for popularity bias in recommenders. Each ingredient is established, but the combination is new and practically useful—no retraining, fine-grained control over the fairness-accuracy trade-off. The paper is clearly written and easy to follow.

The soft spot is the interpretability claim, which is under-validated. The synthetic profiles in Section 3.1 replace every item in a user's history with a random popular or unpopular item. That preserves sequence length but destroys item content, genre, and temporal coherence. SASRec is sequential, so an all-popular history is not just 'popular'—it's also more homogeneous and temporally atypical. Cohen's d then picks neurons that may respond to these correlated features rather than to a general popularity signal. The manipulation study in 5.3.1 reuses the same synthetic-selected neurons, so it doesn't independently confirm the popularity interpretation. The random-noise ablation only shows targeted steering beats random perturbation, not that the target is truly popularity. So the method works, but the explanation of why it works is not nailed down. If popularity correlates differently with content in another domain, the selected neurons may not transfer. That's a limitation, not a fatal flaw.

Minor caveats: no error bars or significance tests, and alpha and N are tuned per dataset, so the reported points are best-case. The figures would be more convincing with error bars or at least multiple runs.

Bottom line: this is a useful contribution to the recommender fairness subfield. It deserves a serious referee. If I were reviewing, I'd ask the authors to validate the interpretability claim—e.g., control for genre/temporal features in the synthetic probes or show that selected neurons transfer across datasets. But the core effectiveness result probably stands.

Recommendation: send to peer review, with a request for that validation.

Referee Report

4 major / 7 minor

Summary. The manuscript proposes PopSteer, a post-hoc interpretability and debiasing method for sequential recommender systems. A sparse autoencoder (SAE) is trained to reconstruct the user embedding produced by a pretrained SASRec model. Synthetic user histories are constructed by replacing every item in a real user's history with a random item from either the popular or the unpopular item set; Cohen's d between SAE activations on the two synthetic sets is used to identify 'popularity neurons.' PopSteer then adjusts the activations of the highest-|d| neurons, suppressing positive-d neurons and boosting negative-d neurons, and the modified user embedding is used to generate recommendations. Experiments on ML-1M and Last.fm compare PopSteer with IPR, FA*IR, PCT, P-MMF, and a random reranker, and an ablation compares targeted steering with random Gaussian activation noise. The paper reports improved long-tail coverage and Gini index with small nDCG@10 loss, and a neuron-deactivation study that links the selected neurons to changes in the Gini index.

Significance. If the interpretability claim is validated, PopSteer would be a practically useful post-hoc tool: it is model-agnostic at the embedding level, offers a fine-grained fairness--accuracy trade-off, and ships with code. The paper also deserves credit for comparing against several established baselines and for including a random-noise ablation, which is a good control for asking whether the steering mechanism itself matters. The central effectiveness result appears plausible. However, the paper's distinctive contribution is the claim that the steered neurons encode popularity specifically, and that claim currently rests on a synthetic-probe construction with a potential content confound. The 'significantly' and 'consistently' statements in the results are also stronger than the reported point estimates support. With additional validation, the method would be a solid contribution to interpretable fairness in recommendation.

major comments (4)

[Section 3.1 and 3.2] The synthetic-probe construction does not isolate popularity from item content and sequence coherence. Replacing every item in a user's history with a randomly selected item from I_pop or I_unpop preserves only the profile length; it destroys genre, artist, recency, and the original temporal ordering, all of which are properties that a sequential model like SASRec can exploit. An all-popular synthetic sequence and an all-unpopular synthetic sequence therefore differ in many dimensions besides popularity. The Cohen's d selection in Eq. (2) may consequently identify neurons that encode 'mainstream content,' 'genre typicality,' or 'sequence predictability' rather than popularity per se. The interpretability analysis in Section 5.3.1 reuses exactly these synthetic-selected neurons, so it does not independently validate the popularity interpretation, and the random-noise ablation in Section 5.3.2 only establishes that targeted selection beats unstructured perturbation. I recommend adding a validation experiment that controls for content, for example by matching genres/categories and recency distributions between the popular and unpopular probe sets, or by testing whether the selected neurons predict popularity on held-out natural user histories.
[Section 5.3.2, Table 2] The claim that 'PopSteer consistently yields superior fairness outcomes relative to random Gaussian noise perturbation' is contradicted by the reported numbers. In ML-1M with N=1024, random noise achieves a lower Gini index (0.7387 vs. 0.7470) and higher long-tail coverage (0.5333 vs. 0.5065) than PopSteer; with N=2048, random noise also has higher long-tail coverage (0.5440 vs. 0.5346). Since lower Gini and higher coverage are better, the noise baseline is at least as fair in those configurations. The comparison is also not accuracy-matched: at N=1024, PopSteer has nDCG@10 of 0.1169 while noise has 0.1123. The ablation should be re-run with matched accuracy (or a proper Pareto-front comparison), and the 'consistently' claim should be tempered.
[Section 4, Eq. (4)] The steering equation uses sigma_j without defining it. Eq. (2) defines sigma_j,pop and sigma_j,unpop separately, but Eq. (4) refers to a single sigma_j. Is this the pooled standard deviation, one of the two group standard deviations, or another quantity? The magnitude of the activation adjustment depends directly on this choice, so the method as written is not fully specified and the implementation is not reproducible without this detail.
[Section 5.2, Figures 1 and 2] The text says PopSteer 'significantly improves fairness' and 'consistently outperforms all baselines,' but no variance estimates, confidence intervals, or significance tests are reported anywhere in the paper. The reported points are single values after per-dataset hyperparameter tuning, which does not support 'significantly' in a statistical sense. At minimum, multiple seeds with error bars and a paired significance test should be reported; otherwise the wording should be weakened to describe the observed improvements in these experiments.

minor comments (7)

[Section 4 and Section 5.2] The symbol N is used both for the total number of SAE hidden neurons and for the number of steered neurons selected by PopSteer, which is confusing. Please use distinct notation, for example N_hidden and N_steer.
[Section 3.1] The phrase 'both compromising roughly 20% of total interactions' should be 'comprising'; more importantly, the definition of the 20% threshold is ambiguous. Please specify whether I_pop and I_unpop are defined by cumulative interaction share, by rank cutoff, or by another rule.
[Section 5.1] There is a typo: 'timestamp informtion' should be 'timestamp information.'
[Section 2] The text writes 'ReLu' where it should be 'ReLU.'
[Section 5.2] The SAE hyperparameters s and K are only given as search ranges; the final chosen values for each dataset are not reported. Please report the settings used in the main experiments.
[Section 5.3.1] The number K' of deactivated neurons in the interpretability analysis is never defined, and Figure 3 does not show numerical values. Please specify the range of K' and, if possible, include the axis values or a small table.
[Eq. (1)] The auxiliary loss L_aux is referenced but not defined in the text; if space is a constraint, a one-sentence description or a precise citation to the formula in [8] would improve reproducibility.

Circularity Check

1 steps flagged · score 4.0 of 10

Interpretability validation is partially self-definitional: neuron labels, steering direction, and fairness metrics all derive from the same I_pop/I_unpop split. The main fairness-accuracy result and noise ablation remain non-circular.

self definitional [Section 3.2 (Eq. 2), Section 4 (Eq. 4), Section 5.3.1 (Fig. 3)]
"To quantify how much each SAE neuron contributes to popularity bias, we compare neuron activations when processing R_pop and R_unpop. ... Positive values suggest alignment with popular content, while negative values imply a focus on unpopular or niche items. ... we manually deactivated the top-K′ neurons identified by PopSteer as most strongly associated with popularity (Cohen's d > 1) or unpopularity (Cohen's d < −1)."

Cohen's d in Eq. 2 is computed as the activation difference between synthetic profiles constructed entirely from I_pop and entirely from I_unpop, where these sets are the same interaction-frequency split used to define the fairness metrics (long-tail coverage and Gini). A neuron is labeled 'popularity-positive' if d>0. The steering rule in Eq. 4 suppresses exactly those d>0 neurons, and the interpretability experiment deactivates the same d>1 neurons and observes a Gini decrease. The direction of the observed effect is therefore set by the definition of the label and the steering sign convention; the experiment is a consistency check of the labeling scheme, not an independent confirmation that the neurons encode a latent popularity feature.

full rationale

The paper's central effectiveness claim—that PopSteer improves long-tail coverage and Gini with minimal nDCG loss—is tested on held-out real user interactions and compared against random Gaussian noise at a matched accuracy budget. This part of the derivation is not circular: the steering targets neurons chosen from synthetic contrasts, yet the fairness improvement is measured on held-out real test interactions, and the random-noise ablation provides a proper control for the value of targeted selection. The only author self-citations ([1], [20], [21]) support standard definitions or metrics that are also backed by independent references, so they are not load-bearing. The main circularity concern is narrower: the interpretability experiment in Section 5.3.1 reuses the same I_pop/I_unpop contrast that defines both the neuron labels and the fairness outcome, making the observed Gini decrease a consistency check rather than an independent validation of the popularity interpretation. This does not invalidate the fairness-accuracy results, but it weakens the paper's interpretability claim. The synthetic-generation confound noted by the skeptic is a validity risk, not a circularity, because the method does not claim to isolate content from popularity in a formal causal sense.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central method introduces no new entities, only a new use of existing SAE and steering machinery. The main free parameters are the steering strength, the number of steered neurons, the SAE sparsity level, and the popularity threshold. The most fragile axioms are the synthetic-profile isolation assumption and the Gaussian activation assumption.

free parameters (4)

alpha (steering strength) = tuned per dataset from {1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0}
Controls the overall intensity of neuron activation adjustment in Eq. 3. The paper tunes it independently for each dataset.
N (number of steered neurons) = tuned per dataset from {1024, 2048, 3072, 4096}
Determines how many neurons with the highest absolute Cohen's d are selected for adjustment. Tuned per dataset.
SAE sparsity K = tuned from {16, 32, 48}
Number of top activations retained in the SAE hidden layer. Affects which neurons are available for steering.
popular/unpopular threshold = roughly 20% of total interactions
Item sets I_pop and I_unpop are defined as the most and least frequently interacted items, following prior work. This cutoff is a modeling choice.

assumptions (4)

domain assumption Activation distributions of neurons approximate a Gaussian, justifying the use of Cohen's d.
Section 3.2 relies on this property to interpret activation differences with Cohen's d, citing prior work on Gaussian pre-activations.
ad hoc to paper Synthetic profiles created by random item replacement isolate the popularity signal while holding all other user characteristics constant.
Section 3.1 assumes that swapping every item with a random popular or unpopular item does not introduce other confounding signals, which is not validated.
domain assumption The SAE's reconstruction of the final user embedding captures the recommender's decision-relevant structure.
Section 2 states that the sparse hidden representation reveals the underlying decision structure, which is an assumption that the embedding is sufficient for interpreting model behavior.
domain assumption Post-hoc modification of the user embedding, followed by the base model's item embeddings, yields valid recommendation scores.
Section 4 assumes that the SAE-modified user embedding remains compatible with the base model's item embedding space, so that the resulting scores are meaningful.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Opening the Black Box: Interpretable Remedies for Popularity Bias in Recommender Systems." pith.science (2026). https://pith.science/paper/DO6PVW24

@misc{pith2026250817297,
  author       = {Pith},
  title        = {Pith review of: Opening the Black Box: Interpretable Remedies for Popularity Bias in Recommender Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DO6PVW24}},
  note         = {Machine review of arXiv:2508.17297}
}

read the original abstract

Popularity bias is a well-known challenge in recommender systems, where a small number of popular items receive disproportionate attention, while the majority of less popular items are largely overlooked. This imbalance often results in reduced recommendation quality and unfair exposure of items. Although existing mitigation techniques address this bias to some extent, they typically lack transparency in how they operate. In this paper, we propose a post-hoc method using a Sparse Autoencoder (SAE) to interpret and mitigate popularity bias in deep recommendation models. The SAE is trained to replicate a pre-trained model's behavior while enabling neuron-level interpretability. By introducing synthetic users with clear preferences for either popular or unpopular items, we identify neurons encoding popularity signals based on their activation patterns. We then adjust the activations of the most biased neurons to steer recommendations toward fairer exposure. Experiments on two public datasets using a sequential recommendation model show that our method significantly improves fairness with minimal impact on accuracy. Moreover, it offers interpretability and fine-grained control over the fairness-accuracy trade-off.

Figures

Figures reproduced from arXiv: 2508.17297 by the authors.

**Figure 1.** Performance comparison of PopSteer method with the baselines in terms of nDCG and fairness metrics on ML-1M. 0.1 0.2 0.3 0.4 0.5 0.6 0.85 0.9 0.95 nDCG Long-tail coverage SASRec Random PCT P-MMF IPR FA*IR PopSteer 0.1 0.2 0.3 0.4 0.5 0.6 0.5 0.6 nDCG Gini Index [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Performance comparison of PopSteer method with the baselines in terms of nDCG and fairness metrics on Last.fm. This adjustment boosts neurons promoting unpopular items (𝑑𝑗 < 0) and suppresses those favoring popular items (𝑑𝑗 > 0). By realigning neuron activations, PopSteer mitigates the overrepresentation of popular items, leading to more balanced exposure across items. The updated SAE then returns the modified use… view at source ↗

**Figure 3.** Interpretability analysis of PopSteer: effect of deactivating 𝐾 ′ identified neurons linked to popularity bias. using linear programming, followed by a reranker that modifies each user’s recommendation list using a modified MMR strategy. Hyperparameters are tuned similar to FA*IR. • P-MMF [28]: A resource allocation algorithm based on dualspace optimization. It dynamically and proportionally adjusts exposure acr… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 14 canonical work pages

[1]

Himan Abdollahpouri, Masoud Mansoury, Robin Burke, Bamshad Mobasher, and Edward Malthouse. 2021. User-centered evaluation of popularity bias in recommender systems. InProceedings of the 29th ACM conference on user modeling, adaptation and personalization. 119–129

2021
[2]

Arda Antikacioglu and R Ravi. 2017. Post processing recommender systems for diversity. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . 707–716

work page 2017
[3]

Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nick Turner, Cem Anil, Carson Denison, Amanda Askell, Robert Lasenby, Yifan Wu, Shauna Kravec, Nicholas Schiefer, Tim Maxwell, Nicholas Joseph, Zac Hatfield-Dodds, Alex Tamkin, Karina Nguyen, Brayden McLean, Josiah E Burke, Tristan Hume, Shan Carter, Tom Henighan, and Ch...

work page 2023
[4]

Rocío Cañamares and Pablo Castells. 2018. Should I follow the crowd? A prob- abilistic analysis of the effectiveness of popularity in recommender systems. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 415–424

work page 2018
[5]

Jiawei Chen, Hande Dong, Yang Qiu, Xiangnan He, Xin Xin, Liang Chen, Guli Lin, and Keping Yang. 2021. AutoDebias: Learning to debias for recommendation. In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 21–30

work page 2021
[6]

Jacob Cohen. 2013. Statistical power analysis for the behavioral sciences. routledge

2013
[7]

Esin Durmus, Alex Tamkin, Jack Clark, Jerry Wei, Jonathan Marcus, Joshua Batson, Kunal Handa, Liane Lovitt, Meg Tong, Miles McCain, Oliver Rausch, Saffron Huang, Sam Bowman, Stuart Ritchie, Tom Henighan, and Deep Ganguli

work page
[8]

Leo Gao, Tom Dupré la Tour, Henk Tillman, Gabriel Goh, Rajan Troll, Alec Radford, Ilya Sutskever, Jan Leike, and Jeffrey Wu. 2024. Scaling and evaluating sparse autoencoders. arXiv preprint arXiv:2406.04093 (2024)

arXiv 2024

Show all 35 references

[9]

Sophie Greenwood, Sudalakshmee Chiniah, and Nikhil Garg. 2024. User-item fairness tradeoffs in recommendations. Advances in Neural Information Processing Systems 37 (2024), 114236–114288

2024
[10]

Muhammad Umair Haider, Hammad Rizwan, Hassan Sajjad, Peizhong Ju, and AB Siddique. 2025. Neurons Speak in Ranges: Breaking Free from Discrete Neuronal Attribution. arXiv preprint arXiv:2502.06809 (2025)

2025 arXiv
[11]

Ruben Härle, Felix Friedrich, Manuel Brack, Björn Deiseroth, Patrick Schramowski, and Kristian Kersting. 2024. SCAR: Sparse Conditioned Au- toencoders for Concept Detection and Steering in LLMs. arXiv preprint arXiv:2411.07122 (2024)

2024 arXiv
[12]

F Maxwell Harper and Joseph A Konstan. 2015. The movielens datasets: History and context. Acm transactions on interactive intelligent systems (tiis) 5, 4 (2015), 1–19

2015
[13]

Ruining He, Wang-Cheng Kang, and Julian McAuley. 2017. Translation-based recommendation. In Proceedings of the eleventh ACM conference on recommender systems. 161–169

2017
[14]

Praveen Hegde. [n. d.]. Effectiveness of Sparse Autoencoder for understanding and removing gender bias in LLMs. InNeurIPS 2024 Workshop on Scientific Methods for Understanding Deep Learning

2024
[15]

Jin Huang, Harrie Oosterhuis, Masoud Mansoury, Herke Van Hoof, and Maarten de Rijke. 2024. Going beyond popularity and positivity bias: Correcting for mul- tifactorial bias in recommender systems. In Proceedings of the 47th International ACM SIGIR Conference on Research and De...

2024
[16]

Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In 2018 IEEE international conference on data mining (ICDM) . IEEE, 197–206

2018
[17]

Michael Lan, Philip Torr, Austin Meek, Ashkan Khakzar, David Krueger, and Fazl Barez. 2024. Sparse autoencoders reveal universal feature spaces across large language models. arXiv preprint arXiv:2410.06981 (2024)

2024 arXiv
[18]

Jaehoon Lee, Yasaman Bahri, Roman Novak, Samuel S Schoenholz, Jeffrey Pen- nington, and Jascha Sohl-Dickstein. 2017. Deep neural networks as gaussian processes. arXiv preprint arXiv:1711.00165 (2017)

2017 arXiv
[19]

Siyi Liu and Yujia Zheng. 2020. Long-tail session-based recommendation. In Proceedings of the 14th ACM conference on recommender systems . 509–514

2020
[20]

Masoud Mansoury. 2022. Understanding and mitigating multi-sided exposure bias in recommender systems. ACM SIGWEB Newsletter 2022, Autumn (2022), 1–4

2022
[21]

Masoud Mansoury, Himan Abdollahpouri, Mykola Pechenizkiy, Bamshad Mobasher, and Robin Burke. 2020. Feedback loop and bias amplification in recommender systems. In Proceedings of the 29th ACM international conference on information & knowledge management . 2145–2148

2020
[22]

Masoud Mansoury, Bamshad Mobasher, and Herke van Hoof. 2024. Mitigating exposure bias in online learning to rank recommendation: A novel reward model for cascading bandits. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management . 1638–1648

2024
[23]

Charles O’Neill, Christine Ye, Kartheik Iyer, and John F Wu. 2024. Disentangling dense embeddings with sparse autoencoders. arXiv preprint arXiv:2408.00657 (2024)

2024 arXiv
[24]

Markus Schedl. 2016. The lfm-1b dataset for music retrieval and recommendation. In Proceedings of the 2016 ACM on international conference on multimedia retrieval . 103–110

2016
[25]

Tobias Schnabel, Adith Swaminathan, Ashudeep Singh, Navin Chandak, and Thorsten Joachims. 2016. Recommendations as treatments: Debiasing learning and evaluation. In international conference on machine learning . PMLR, 1670– 1679

2016
[26]

Chenyang Wang, Yankai Liu, Yuanqing Yu, Weizhi Ma, Min Zhang, Yiqun Liu, Haitao Zeng, Junlan Feng, and Chao Deng. 2023. Two-sided calibration for quality- aware responsible recommendation. In Proceedings of the 17th ACM Conference on Recommender Systems. 223–233

2023
[27]

Pierre Wolinski and Julyan Arbel. 2022. Gaussian pre-activations in neural networks: Myth or reality? arXiv preprint arXiv:2205.12379 (2022)

2022 arXiv
[28]

Chen Xu, Sirui Chen, Jun Xu, Weiran Shen, Xiao Zhang, Gang Wang, and Zhenhua Dong. 2023. P-MMF: Provider max-min fairness re-ranking in recommender system. In Proceedings of the ACM Web Conference 2023 . 3701–3711

2023
[29]

Meike Zehlike, Francesco Bonchi, Carlos Castillo, Sara Hajian, Mohamed Mega- hed, and Ricardo Baeza-Yates. 2017. Fa* ir: A fair top-k ranking algorithm. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Man- agement. 1569–1578

2017
[30]

Mi Zhang and Neil Hurley. 2010. Niche product retrieval in top-n recommen- dation. In 2010 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology, Vol. 1. IEEE, 74–81

2010
[31]

Yin Zhang, Ruoxi Wang, Derek Zhiyuan Cheng, Tiansheng Yao, Xinyang Yi, Lichan Hong, James Caverlee, and Ed H Chi. 2023. Empowering long-tail item recommendation through cross decoupling network (CDN). In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and ...

2023
[32]

Ziwei Zhu and James Caverlee. 2022. Fighting mainstream bias in recommender systems via local fine tuning. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining . 1497–1506

2022
[33]

Ziwei Zhu, Yun He, Xing Zhao, Yin Zhang, Jianling Wang, and James Caverlee
[2021]

In Proceedings of the 14th ACM international conference on web search and data mining

Popularity-opportunity bias in collaborative filtering. In Proceedings of the 14th ACM international conference on web search and data mining . 85–93
[2024]

https: //anthropic.com/research/evaluating-feature-steering

Evaluating Feature Steering: A Case Study in Mitigating Social Biases . https: //anthropic.com/research/evaluating-feature-steering

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Himan Abdollahpouri, Masoud Mansoury, Robin Burke, Bamshad Mobasher, and Edward Malthouse. 2021. User-centered evaluation of popularity bias in recommender systems. InProceedings of the 29th ACM conference on user modeling, adaptation and personalization. 119–129

2021

[2] [2]

Arda Antikacioglu and R Ravi. 2017. Post processing recommender systems for diversity. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . 707–716

work page 2017

[3] [3]

Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nick Turner, Cem Anil, Carson Denison, Amanda Askell, Robert Lasenby, Yifan Wu, Shauna Kravec, Nicholas Schiefer, Tim Maxwell, Nicholas Joseph, Zac Hatfield-Dodds, Alex Tamkin, Karina Nguyen, Brayden McLean, Josiah E Burke, Tristan Hume, Shan Carter, Tom Henighan, and Ch...

work page 2023

[4] [4]

Rocío Cañamares and Pablo Castells. 2018. Should I follow the crowd? A prob- abilistic analysis of the effectiveness of popularity in recommender systems. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 415–424

work page 2018

[5] [5]

Jiawei Chen, Hande Dong, Yang Qiu, Xiangnan He, Xin Xin, Liang Chen, Guli Lin, and Keping Yang. 2021. AutoDebias: Learning to debias for recommendation. In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 21–30

work page 2021

[6] [6]

Jacob Cohen. 2013. Statistical power analysis for the behavioral sciences. routledge

2013

[7] [7]

Esin Durmus, Alex Tamkin, Jack Clark, Jerry Wei, Jonathan Marcus, Joshua Batson, Kunal Handa, Liane Lovitt, Meg Tong, Miles McCain, Oliver Rausch, Saffron Huang, Sam Bowman, Stuart Ritchie, Tom Henighan, and Deep Ganguli

work page

[8] [8]

Leo Gao, Tom Dupré la Tour, Henk Tillman, Gabriel Goh, Rajan Troll, Alec Radford, Ilya Sutskever, Jan Leike, and Jeffrey Wu. 2024. Scaling and evaluating sparse autoencoders. arXiv preprint arXiv:2406.04093 (2024)

arXiv 2024