Pith. sign in

REVIEW 2 major objections 4 minor 37 references

A post-hoc reweighting step can force recommender outputs toward any chosen category mix without retraining the model.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 08:18 UTC pith:PIJPEZUZ

load-bearing objection Clean, usable post-hoc reweighting that turns classical label shift into a stakeholder-level control for attribute exposure; solid engineering contribution with one disclosed approximation. the 2 major comments →

arxiv 2607.10915 v1 pith:PIJPEZUZ submitted 2026-07-12 cs.IR cs.LG

Normative Alignment of Recommender Systems via Internal Label Shift

Johannes Kruse , Kasper Lindskow , Michael Riis Andersen , Ryotaro Shimizu , Julian McAuley , Pierre-Alexandre Mattei , Jes Frellsen This is my paper
classification cs.IR cs.LG
keywords Recommender SystemsAligned RecommendationNormative DesignLabel ShiftRelevance Prioritized RerankingCalibrationNews Recommendation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Recommenders that chase clicks alone can drift away from the category mix an organization actually wants—fairness targets, editorial balance, or equal airtime across topics. This paper shows that you can correct that drift after the model is already trained. The method, NAILS, treats the desired category proportions as a label-shift problem inside a hierarchical item–attribute model, multiplies each item’s score by a simple ratio of target to observed category mass, and re-normalizes. Because the correction sits on top of an existing recommender, no new training is required and personalization is largely preserved. On a large news-recommendation dataset the adjusted lists move measurably closer to both editorial and uniform category targets while AUC either stays flat or even rises for some targets. The result is a practical dial that platforms can turn to keep recommendations inside their normative bounds.

Core claim

NAILS can reweight an already-trained recommender’s user-conditional item scores so that the induced marginal distribution over item attributes matches a stakeholder-chosen target, while leaving the learned preference structure essentially intact and without any model retraining.

What carries the argument

Internal label shift: the classical Saerens-style ratio of target to source attribute probabilities is applied at the internal (attribute) level of a hierarchical item–attribute model, then summed over attributes and re-normalized, producing a corrected item distribution that can be mixed with the original via a single scalar λ.

Load-bearing premise

The method assumes that how users and items behave given a category stays the same when only the overall mix of categories is changed, and that any resulting change in the user population can be ignored.

What would settle it

Measure the realized category distribution of top-K recommendations after applying NAILS on a held-out user population whose true category-conditional preferences differ from those used to estimate the correction; if the observed mix fails to approach the target while engagement collapses, the claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Platforms can enforce global editorial or fairness targets on live traffic without retraining or replacing their existing recommender.
  • A single scalar λ lets operators trade alignment strength against ranking metrics in real time.
  • The same post-hoc correction applies to any base model that emits item scores, making it model-agnostic.
  • Choice of target distribution itself becomes a design decision that can improve or degrade engagement, so targets must be chosen carefully.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the correction is a simple multiplicative reweight, it can be pushed into the serving path as a cheap, cacheable feature rather than a heavy re-ranker.
  • Longitudinal A/B tests would reveal whether the short-term AUC preservation survives once users adapt to the new category mix.
  • The same internal-shift idea could be applied to multi-label attributes or continuous quality scores once the hierarchical model is generalized.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper proposes NAILS, a post-hoc reweighting method that aligns a recommender’s user-conditional item distribution with a stakeholder-specified marginal over item attributes (e.g., news categories) by treating the problem as internal label shift inside a hierarchical model. Starting from the joint p(c,i,u)=p(c|i)p_θ(i|u)p(u), it derives a corrected ˜p(i|u) that multiplies the original scores by a category-dependent factor involving the ratio ˜p(c)/p(c), optionally mixed by a scalar λ. The method requires no retraining, is model-agnostic, and is evaluated on the public EB-NeRD news dataset with NRMS as the base model against the greedy CaliRec baseline, reporting KL divergence of top-10 category distributions, coverage, and AUC under editorial and uniform targets.

Significance. If the empirical claims hold, NAILS supplies a lightweight, scalable, and theoretically transparent mechanism for stakeholder-centric (rather than purely user-centric) normative alignment. The derivation is classical label-shift applied at an internal node of a hierarchy, the approximation of freezing p(u) is explicitly acknowledged, and the public code plus reproducible EB-NeRD experiments make the contribution immediately usable by news platforms and other multi-stakeholder recommenders. The work therefore advances the practical toolkit for value-driven recommendation without requiring architectural changes or expensive re-optimization.

major comments (2)
  1. Section 5 and Figures 1–2 report only point estimates of KL and AUC; no standard errors, bootstrap intervals, or statistical tests are provided. Given that the central claim is “minimal impact on user engagement,” the absence of uncertainty quantification makes it impossible to judge whether the observed AUC differences (especially the modest gains under the editorial target) are reliable or simply sampling noise on the 13 M impression test set.
  2. All ranking and calibration results rest on a single base model (NRMS). While the method is presented as model-agnostic, the load-bearing empirical claim that alignment “consistently improves attribute-level alignment with minimal impact” has not been stress-tested on any other architecture (e.g., sequential or graph-based recommenders). A second base model would substantially strengthen the generality argument.
minor comments (4)
  1. After Equation (5) the authors correctly note that freezing p(u) yields only an approximate solution; a short quantitative diagnostic (e.g., the KL between the induced ˜p(c) and the target under the frozen-p(u) model) would help readers gauge the size of the approximation error on the data used.
  2. Table 1 lists KL@10 for NAILS-stoch that remain high (~2.0) under the editorial target even at large λ; a brief discussion of why stochastic sampling fails to match the deterministic and CaliRec numbers would clarify the practical trade-off.
  3. The hyper-parameter search for NRMS is described, yet the final λ values chosen for the “best” configurations in Figure 1 are not justified by a formal selection criterion (validation KL vs. AUC Pareto front, for example).
  4. Minor notation: p(c|i) is introduced as a general probabilistic mapping but is immediately specialized to an indicator; stating the deterministic case up front would avoid a momentary ambiguity.

Circularity Check

0 steps flagged

No significant circularity: classical label-shift reweighting applied post-hoc, with independent held-out evaluation

full rationale

The derivation in §3 starts from the joint p(c,i,u)=p(c|i)p_θ(i|u)p(u), imposes the external Saerens et al. (2002) invariance p(i,u|c)=~p(i,u|c), and obtains the explicit reweighting factor ∑_c (~p(c)/p(c))p(c|i) that multiplies the original scores (Eqs. 3–5). The only approximation (freezing p(u)) is stated openly and does not make the target distribution tautological. Target distributions ~p(c) are defined independently of the evaluation metrics (editorial frequencies from candidate lists; uniform). p(c) is estimated once from the base model and never fitted to the KL or AUC numbers reported on held-out impressions. Lambda is a free mixing parameter whose effect is shown, not optimized against the claimed metrics. No self-citation supplies a uniqueness theorem or ansatz that forces the result; the method is a transparent, model-agnostic post-processing step whose empirical success is measured rather than constructed. Hence the central claim of improved attribute alignment with modest engagement cost does not reduce to its inputs by definition.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 0 invented entities

The method rests on the classical label-shift invariance assumption plus a practical approximation that freezes the user marginal; the only free parameter is the mixing weight lambda. No new physical or mathematical entities are invented.

free parameters (1)
  • lambda = tuned in [0,1]; best values reported per target (e.g. 0.7 editorial, 0.3/0.99 uniform)
    Mixing coefficient in Eq. 6 that interpolates between original and fully-aligned scores; chosen by the operator (or by Optuna-style search) and directly controls the strength of the reported alignment.
axioms (3)
  • domain assumption p(i,u|c) is invariant under a change of the attribute marginal p(c) (classical label-shift assumption)
    Invoked immediately after Eq. 2 to obtain the reweighting formula; taken from Saerens et al. 2002 without further justification for the recommender setting.
  • ad hoc to paper The induced shift in the user marginal can be ignored by keeping p(u) fixed
    Explicitly stated after Eq. 5; the authors note that the corrected model therefore only approximately matches the target p̃(c).
  • domain assumption Each item belongs to exactly one category so p(c|i) is an indicator
    Stated in §4; simplifies the hierarchical model but is not required by the general derivation.

pith-pipeline@v1.1.0-grok45 · 15824 in / 2095 out tokens · 21274 ms · 2026-07-14T08:18:41.490995+00:00 · methodology

0 comments
read the original abstract

We introduce NAILS (Normative Alignment of Recommender Systems via Internal Label Shift), a simple and scalable method for aligning recommendation outputs with target distributions over item-level attributes, such as categories. Recommender systems optimized solely for user engagement often fail to satisfy broader normative objectives, including fairness, diversity, and editorial values. NAILS modifies the user-conditional item distribution to induce a specified marginal distribution over attributes while preserving the preferences learned by an existing recommender system and requiring no model retraining. We formulate this problem as a form of label shift applied internally within a hierarchical classification framework. By adopting a stakeholder-centric perspective, NAILS enables recommendation outputs to be aligned with global normative objectives. Empirically, we show that NAILS consistently improves attribute-level alignment with minimal impact on user engagement, providing a practical mechanism for value-driven recommendation.

Figures

Figures reproduced from arXiv: 2607.10915 by Jes Frellsen, Johannes Kruse, Julian McAuley, Kasper Lindskow, Michael Riis Andersen, Pierre-Alexandre Mattei, Ryotaro Shimizu.

Figure 2
Figure 2. Figure 2: AUC (↑) across varying calibration strengths (𝜆). Larger markers indicate the best performance. deterministic methods, as our goal is to assess the effect of align￾ment in a stable and reproducible setting [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Category distribution among the top-10 recom [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 11 canonical work pages

  1. [1]

    Himan Abdollahpouri, Zahra Nazari, Alex Gain, Clay Gibson, Maria Di- makopoulou, Jesse Anderton, Benjamin Carterette, Mounia Lalmas, and Tony Jebara. 2023. Calibrated Recommendations as a Minimum-Cost Flow Problem. InProceedings of the Sixteenth ACM International Conference on Web Search and Data Mining(Singapore, Singapore)(WSDM ’23). Association for Com...

    doi:10.1145/3539597.3570402 2023

  2. [2]

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. 2019. Optuna: A Next-generation Hyperparameter Optimization Framework. InProceedings of the 25th ACM SIGKDD International Confer- ence on Knowledge Discovery & Data Mining(Anchorage, AK, USA)(KDD ’19). Association for Computing Machinery, New York, NY, USA, 2623–2631. https://...

    doi:10.1145/3292500.3330701 2019

  3. [3]

    Jiayi Chen, Wen Wu, Liye Shi, Yu Ji, Wenxin Hu, Xi Chen, Wei Zheng, and Liang He. 2022. DACSR: Decoupled-Aggregated End-to-End Calibrated Sequen- tial Recommendation.Applied Sciences12, 22 (2022). https://doi.org/10.3390/ app122211765

    2022

  4. [4]

    Yashar Deldjoo, Fatemeh Nazary, Arnau Ramisa, Julian McAuley, Giovanni Pel- legrini, Alejandro Bellogin, and Tommaso Di Noia. 2023. A Review of Modern Fashion Recommender Systems.ACM Comput. Surv.56, 4, Article 87 (Oct. 2023), 37 pages. https://doi.org/10.1145/3624733

    work page doi:10.1145/3624733 2023

  5. [5]

    Saurabh Garg, Yifan Wu, Sivaraman Balakrishnan, and Zachary Lipton. 2020. A unified view of label shift estimation.Advances in Neural Information Processing Systems33 (2020), 3290–3300

    2020

  6. [6]

    Mouzhi Ge, Carla Delgado-Battenfeld, and Dietmar Jannach. 2010. Beyond Accuracy: Evaluating Recommender Systems by Coverage and Serendipity. In Proceedings of the Fourth ACM Conference on Recommender Systems(Barcelona, Spain)(RecSys ’10). Association for Computing Machinery, New York, NY, USA, 257–260. https://doi.org/10.1145/1864708.1864761

    doi:10.1145/1864708.1864761 2010

  7. [7]

    Natali Helberger. 2019. On the Democratic Role of News Recommenders.Digital Journalism7, 8 (2019), 993–1012

    2019

  8. [8]

    Andreea Iana, Goran Glavaš, and Heiko Paulheim. 2024. Train Once, Use Flexibly: A Modular Framework for Multi-Aspect Neural News Recommen- dation. InFindings of the Association for Computational Linguistics: EMNLP 2024, Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen (Eds.). Associa- tion for Computational Linguistics, Miami, Florida, USA, 9555–9571. ht...

    work page doi:10.18653/v1/2024.findings-emnlp.558 2024

  9. [9]

    Dietmar Jannach and Michael Jugovac. 2019. Measuring the Business Value of Recommender Systems.ACM Trans. Manage. Inf. Syst.10, 4, Article 16 (dec 2019), 23 pages. https://doi.org/10.1145/3370082

    doi:10.1145/3370082 2019

  10. [10]

    Hyunsik Jeon, Se-eun Yoon, and Julian McAuley. 2024. Calibration-Disentangled Learning and Relevance-Prioritized Reranking for Calibrated Sequential Recom- mendation. InProceedings of the 33rd ACM International Conference on Informa- tion and Knowledge Management(Boise, ID, USA)(CIKM ’24). Association for Computing Machinery, New York, NY, USA, 973–982. h...

    arXiv 2024

  11. [11]

    Marius Kaminskas and Derek Bridge. 2016. Diversity, Serendipity, Novelty, and Coverage: A Survey and Empirical Analysis of Beyond-Accuracy Objectives in Recommender Systems.ACM Trans. Interact. Intell. Syst.7, 1, Article 2 (dec 2016), 42 pages

    2016

  12. [12]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2017. Adam: A Method for Stochastic Opti- mization. arXiv:1412.6980 [cs.LG] https://arxiv.org/abs/1412.6980

    Pith/arXiv arXiv 2017

  13. [13]

    Daphne Koller and Mehran Sahami. 1997. Hierarchically Classifying Documents Using Very Few Words. InProceedings of the Fourteenth International Conference on Machine Learning (ICML ’97). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 170–178

    1997

  14. [14]

    Johannes Kruse, Kasper Lindskow, Michael Riis Andersen, and Jes Frellsen. 2023. Creating the next generation of news experience on ekstrabladet.dk with rec- ommender systems. InProceedings of the 17th ACM Conference on Recommender Systems(Singapore, Singapore)(RecSys ’23). Association for Computing Machin- ery, New York, NY, USA, 1067–1070. https://doi.or...

    work page doi:10.1145/3604915.3610248 2023

  15. [15]

    Johannes Kruse, Kasper Lindskow, Saikishore Kalloori, Marco Polignano, Clau- dio Pomo, Abhishek Srivastava, Anshuk Uppal, Michael Riis Andersen, and Jes Frellsen. 2024. EB-NeRD a large-scale dataset for news recommendation. In Proceedings of the Recommender Systems Challenge 2024(Bari, Italy)(RecSysChal- lenge ’24). Association for Computing Machinery, Ne...

    work page doi:10.1145/3687151.3687152 2024

  16. [16]

    Johannes Kruse, Kasper Lindskow, Saikishore Kalloori, Marco Polignano, Clau- dio Pomo, Abhishek Srivastava, Anshuk Uppal, Michael Riis Andersen, and Jes Frellsen. 2024. RecSys Challenge 2024: Balancing Accuracy and Editorial Values in News Recommendations. InProceedings of the 18th ACM Conference on Recom- mender Systems(Bari, Italy)(RecSys ’24). Associat...

    work page doi:10.1145/3640457.3687164 2024

  17. [17]

    Jian Li, Jieming Zhu, Qiwei Bi, Guohao Cai, Lifeng Shang, Zhenhua Dong, Xin Jiang, and Qun Liu. 2022. MINER: Multi-Interest Matching Network for News Recommendation. InFindings of the Association for Computational Linguistics: ACL 2022, Smaranda Muresan, Preslav Nakov, and Aline Villavicencio (Eds.). Association for Computational Linguistics, Dublin, Irel...

    work page doi:10.18653/v1/2022.findings-acl.29 2022

  18. [18]

    Feng Lu, Anca Dumitrache, and David Graus. 2020. Beyond Optimizing for Clicks: Incorporating Editorial Values in News Recommendation. InProceedings of the 28th ACM Conference on User Modeling, Adaptation and Personalization (Genoa, Italy)(UMAP ’20). Association for Computing Machinery, New York, NY, USA, 145–153. https://doi.org/10.1145/3340631.3394864

    doi:10.1145/3340631.3394864 2020

  19. [19]

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arXiv:1301.3781 [cs.CL] https://arxiv.org/abs/1301.3781

    Pith/arXiv arXiv 2013

  20. [20]

    Efrat Nechushtai and Seth C. Lewis. 2019. What kind of news gatekeepers do we want machines to be? Filter bubbles, fragmentation, and the normative dimensions of algorithmic recommendations.Computers in Human Behavior90 (2019), 298–307. https://doi.org/10.1016/j.chb.2018.07.043

    work page doi:10.1016/j.chb.2018.07.043 2019

  21. [21]

    Tao Qi, Fangzhao Wu, Chuhan Wu, and Yongfeng Huang. 2022. News Rec- ommendation with Candidate-aware User Modeling. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval(Madrid, Spain)(SIGIR ’22). Association for Computing Machinery, New York, NY, USA, 1917–1921. https://doi.org/10.1145/3477495.3531778

    work page doi:10.1145/3477495.3531778 2022

  22. [22]

    Marco Saerens, Patrice Latinne, and Christine Decaestecker. 2002. Adjusting the outputs of a classifier to new a priori probabilities: a simple procedure.Neural computation14, 1 (2002), 21–41

    2002

  23. [23]

    Malthouse

    Sinan Seymen, Himan Abdollahpouri, and Edward C. Malthouse. 2021. A Con- strained Optimization Approach for Calibrated Recommendations. InProceedings of the 15th ACM Conference on Recommender Systems(Amsterdam, Netherlands) (RecSys ’21). Association for Computing Machinery, New York, NY, USA, 607–612. https://doi.org/10.1145/3460231.3478857

    work page doi:10.1145/3460231.3478857 2021

  24. [24]

    Ryotaro Shimizu, Takashi Wada, Yu Wang, Johannes Kruse, Sean O’Brien, Sai HtaungKham, Linxin Song, Yuya Yoshikawa, Yuki Saito, Fugee Tsung, Masayuki Goto, and Julian McAuley. 2025. Disentangling Likes and Dislikes in Personalized Generative Explainable Recommendation. InProceedings of the ACM on Web Conference 2025(Sydney NSW, Australia)(WWW ’25). Associa...

    doi:10.1145/3696410 2025

  25. [25]

    Ryotaro Shimizu, Yu Wang, Masanari Kimura, Yuki Hirakawa, Takashi Wada, Yuki Saito, and Julian McAuley. 2024. A Fashion Item Recommendation Model in Hyperbolic Space. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. IEEE Computer Society, Seattle, WA, USA, 8377–8383

    2024

  26. [26]

    Silla and Alex A

    Carlos N. Silla and Alex A. Freitas. 2011. A survey of hierarchical classification across different application domains.Data Mining and Knowledge Discovery22, 1 (2011), 31–72

    2011

  27. [27]

    Harald Steck. 2018. Calibrated recommendations. InProceedings of the 12th ACM Conference on Recommender Systems(Vancouver, British Columbia, Canada) (RecSys ’18). Association for Computing Machinery, New York, NY, USA, 154–162. https://doi.org/10.1145/3240323.3240372

    doi:10.1145/3240323.3240372 2018

  28. [28]

    Jonathan Stray, Ivan Vendrov, Jeremy Nixon, Steven Adler, and Dylan Hadfield- Menell. 2021. What are you optimizing for? Aligning Recommender Systems with Human Values. arXiv:2107.10939 [cs.IR] https://arxiv.org/abs/2107.10939

    Pith/arXiv arXiv 2021

  29. [29]

    Sanne Vrijenhoek, Gabriel Bénédict, Mateo Gutierrez Granada, Daan Odijk, and Maarten De Rijke. 2022. RADio – Rank-Aware Divergence Metrics to Measure Normative Diversity in News Recommendations. InProceedings of the 16th ACM Conference on Recommender Systems(Seattle, WA, USA)(RecSys ’22). Association for Computing Machinery, New York, NY, USA, 208–219. ht...

    arXiv 2022

  30. [30]

    Sanne Vrijenhoek, Mesut Kaya, Nadia Metoui, Judith Möller, Daan Odijk, and Natali Helberger. 2021. Recommenders with a Mission: Assessing Diversity in News Recommendations. InProceedings of the 2021 Conference on Human Information Interaction and Retrieval(Canberra ACT, Australia)(CHIIR ’21). Association for Computing Machinery, New York, NY, USA, 173–183...

    doi:10.1145/3406522.3446019 2021

  31. [31]

    Sanne Vrijenhoek, Lien Michiels, Johannes Kruse, Alain Starke, Nava Tintarev, and Jordi Viader Guerrero. 2023. NORMalize: The First Workshop on Normative Design and Evaluation of Recommender Systems. InProceedings of the 17th ACM Conference on Recommender Systems(Singapore, Singapore)(RecSys ’23). Association for Computing Machinery, New York, NY, USA, 12...

    work page doi:10.1145/3604915.3608757 2023

  32. [32]

    Chenyang Wang, Yankai Liu, Yuanqing Yu, Weizhi Ma, Min Zhang, Yiqun Liu, Haitao Zeng, Junlan Feng, and Chao Deng. 2023. Two-sided Calibra- tion for Quality-aware Responsible Recommendation. InProceedings of the 17th ACM Conference on Recommender Systems(Singapore, Singapore)(Rec- Sys ’23). Association for Computing Machinery, New York, NY, USA, 223–233. h...

    doi:10.1145/3604915.3608799 2023

  33. [33]

    Shuhei Watanabe. 2023. Tree-Structured Parzen Estimator: Understanding Its Algorithm Components and Their Roles for Better Empirical Performance. arXiv:2304.11127 [cs.LG] https://arxiv.org/abs/2304.11127

    Pith/arXiv arXiv 2023

  34. [34]

    Chuhan Wu, Fangzhao Wu, Suyu Ge, Tao Qi, Yongfeng Huang, and Xing Xie. 2019. Neural News Recommendation with Multi-Head Self-Attention. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP- IJCNLP). Association for Computational Linguistics...

    2019

  35. [35]

    https://doi.org/10.18653/v1/D19-1671

    work page doi:10.18653/v1/d19-1671

  36. [36]

    Fangzhao Wu, Ying Qiao, Jiun-Hung Chen, Chuhan Wu, Tao Qi, Jianxun Lian, Danyang Liu, Xing Xie, Jianfeng Gao, Winnie Wu, and Ming Zhou. 2020. MIND: A Large-scale Dataset for News Recommendation. InProceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, Online, 3597–3606

    2020

  37. [37]

    Xing Zhao, Ziwei Zhu, Majid Alfifi, and James Caverlee. 2020. Addressing the Target Customer Distortion Problem in Recommender Systems. InPro- ceedings of The Web Conference 2020(Taipei, Taiwan)(WWW ’20). Associ- ation for Computing Machinery, New York, NY, USA, 2969–2975. https: //doi.org/10.1145/3366423.3380065

    doi:10.1145/3366423.3380065 2020