pith. sign in

arxiv: 2606.07492 · v1 · pith:FYZ7LPVWnew · submitted 2026-06-05 · 💻 cs.IR · cs.LG· stat.ML

Bradley-Terry Rankings for Recommender Systems Across Dataset Taxonomies

Ekaterina Grishina , Stepan Kuznetsov , Askar Tsyganov , Ilya Ivanov , Daria Korovaitceva , Margarita Rusanova , Uliana Parkina , Alexander Derevyagin
show 3 more authors
Evgeny Frolov Sergey Samsonov Anton Lysenko
This is my paper

Pith reviewed 2026-06-27 20:24 UTC · model grok-4.3

classification 💻 cs.IR cs.LGstat.ML
keywords Bradley-Terry modelrecommender systemsalgorithm rankingdataset characteristicsranking consistencyunseen datasetspairwise comparisons
0
0 comments X

The pith

Bradley-Terry models produce dataset-aware rankings of recommendation algorithms that generalize to new data.

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

Recommendation algorithm performance varies sharply with dataset properties such as sparsity, sequential structure, and scale, so simply averaging metrics like NDCG across benchmarks often produces misleading orderings. The paper models pairwise wins between algorithms using the Bradley-Terry framework so that the resulting ranking explicitly incorporates those dataset statistics. It also supplies a consistency metric, shows the rankings remain stable under missing comparisons, and extends the model with trees and covariates to predict algorithm orderings on entirely new datasets from statistics alone. A reader would care because the method replaces ad-hoc aggregation with a data-driven procedure that can guide algorithm choice without exhaustive testing.

Core claim

The ranking of recommendation algorithms obtained via the Bradley-Terry model depends on key dataset statistics, and extensions of the framework allow ranking algorithms on unseen datasets without executing the models.

What carries the argument

Bradley-Terry model applied to pairwise algorithm performance comparisons, extended via BT trees and models with covariates to incorporate dataset statistics.

If this is right

  • Rankings of algorithms become sensitive to measurable dataset characteristics rather than remaining uniform across benchmarks.
  • A dedicated metric quantifies how consistent the derived rankings are across different data subsets.
  • The procedure tolerates incomplete pairwise comparison data without large changes in the final order.
  • Algorithm rankings for previously unseen datasets can be produced from dataset statistics alone using BT trees or covariate models.

Where Pith is reading between the lines

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

  • Dataset taxonomies could be built directly from the statistics that most strongly shift the BT rankings.
  • Selection of models for production systems might shift from running full benchmarks to computing a small set of dataset descriptors.
  • The same pairwise modeling approach could be tested on ranking tasks outside recommender systems, such as classifier selection.

Load-bearing premise

Pairwise comparisons of algorithm performance on datasets can be validly modeled by the Bradley-Terry framework and the chosen dataset statistics capture the factors driving performance differences.

What would settle it

A collection of new datasets where the BT-predicted ranking order disagrees with the order obtained by actually running every algorithm and measuring performance.

Figures

Figures reproduced from arXiv: 2606.07492 by Alexander Derevyagin, Anton Lysenko, Askar Tsyganov, Daria Korovaitceva, Ekaterina Grishina, Evgeny Frolov, Ilya Ivanov, Margarita Rusanova, Sergey Samsonov, Stepan Kuznetsov, Uliana Parkina.

Figure 1
Figure 1. Figure 1: Ratio of transitive triples in rankings of different [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Weights of rankings by NDCG@10 for data with [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of weights of different models. Metric [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pairwise performance comparison of recommender models. Each cell [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of weights of different models on [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: BT trees for datasets. Each node displays the split [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

The ranking of recommendation algorithms is a challenging problem since model performance is sensitive to dataset characteristics such as sparsity, sequential structure, and scale. This drives a demand for a proper methodology for fair comparison between algorithms. Naive aggregation of performance metrics (e.g., averaging NDCG over benchmarks) can yield misleading rankings, undermining practical selection. To address this problem, we introduce a novel, data-driven ranking methodology based on Bradley-Terry (BT) model. We demonstrate that the obtained ranking depends on key dataset statistics. Additionally, we propose a novel metric for evaluating ranking consistency and demonstrate robustness of our ranking to incomplete data. Finally, we introduce a dataset-specific methodology for ranking algorithms on unseen datasets without running the models, relying on extensions of the Bradley-Terry framework, including BT trees and BT models with covariates.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces a Bradley-Terry (BT) model-based methodology for ranking recommender system algorithms that accounts for dataset characteristics such as sparsity, sequential structure, and scale. It claims that the derived rankings depend on these key statistics, proposes a novel metric for evaluating ranking consistency, demonstrates robustness to incomplete data, and extends the BT framework (via BT trees and models with covariates) to enable ranking algorithms on unseen datasets without executing the models.

Significance. If the central claims hold, the work could provide a principled alternative to naive metric aggregation for algorithm comparison across heterogeneous datasets, with practical value for selecting models on new data. The extension to covariate-adjusted BT models for zero-shot ranking on unseen datasets is potentially impactful if the chosen statistics prove sufficient, but this requires strong empirical support that is not verifiable from the abstract alone.

major comments (2)
  1. [Section describing BT models with covariates / unseen dataset methodology] The load-bearing claim for the dataset-specific methodology on unseen datasets (BT models with covariates and BT trees) requires that the selected dataset statistics suffice as covariates to capture performance drivers. The manuscript must include ablation or sensitivity analyses showing that predictions generalize beyond the fitted data and are not undermined by unmodeled factors (e.g., domain-specific distributions). Without this, the extension reduces to an unvalidated assumption.
  2. [Results section on ranking dependence on dataset statistics] The demonstration that obtained rankings depend on key dataset statistics needs explicit quantification (e.g., via regression coefficients or ranking shifts in tables) rather than qualitative statements; otherwise the dependence claim lacks load-bearing evidence.
minor comments (2)
  1. [Metric definition] Clarify the exact form of the consistency metric and how it is computed from pairwise BT outcomes.
  2. [Covariate selection] Provide the full set of dataset statistics used as covariates and justify their selection with references to prior recsys literature on performance factors.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the empirical support and clarity of our claims.

read point-by-point responses

  1. Referee: [Section describing BT models with covariates / unseen dataset methodology] The load-bearing claim for the dataset-specific methodology on unseen datasets (BT models with covariates and BT trees) requires that the selected dataset statistics suffice as covariates to capture performance drivers. The manuscript must include ablation or sensitivity analyses showing that predictions generalize beyond the fitted data and are not undermined by unmodeled factors (e.g., domain-specific distributions). Without this, the extension reduces to an unvalidated assumption.

    Authors: We agree that the generalization capability of the BT models with covariates and BT trees requires stronger empirical validation to confirm that the chosen dataset statistics adequately capture performance drivers. In the revised manuscript, we will add ablation studies that systematically remove or perturb individual covariates and measure out-of-sample prediction accuracy on held-out datasets. We will also include sensitivity analyses that evaluate prediction performance across different data domains to quantify the effect of potential unmodeled factors. These additions will directly address the concern and provide the necessary evidence for the zero-shot ranking methodology. revision: yes

  2. Referee: [Results section on ranking dependence on dataset statistics] The demonstration that obtained rankings depend on key dataset statistics needs explicit quantification (e.g., via regression coefficients or ranking shifts in tables) rather than qualitative statements; otherwise the dependence claim lacks load-bearing evidence.

    Authors: We acknowledge that the current description of ranking dependence on dataset statistics is primarily qualitative. To provide explicit, load-bearing evidence, we will revise the results section to include tables reporting the estimated coefficients from the Bradley-Terry models for each key statistic (sparsity, sequential structure, and scale). We will also add quantitative examples of ranking shifts observed when varying these statistics, such as changes in algorithm positions across different sparsity levels. This will make the dependence claim more rigorous and verifiable. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard BT application with independent statistical extensions

full rationale

The provided abstract and claims describe application of the established Bradley-Terry model to derive rankings from pairwise algorithm comparisons on datasets, with dependence on statistics like sparsity shown empirically rather than by definitional reduction. Extensions to BT trees and covariate models are standard and not shown to reduce to fitted parameters by construction or via self-citation chains. No equations, self-citations, or ansatzes are quoted that force the central claims (e.g., unseen-dataset prediction) to equal their inputs. The derivation chain is therefore self-contained against external benchmarks like the BT literature.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; ledger left empty.

pith-pipeline@v0.9.1-grok · 5716 in / 954 out tokens · 17051 ms · 2026-06-27T20:24:04.913428+00:00 · methodology

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

51 extracted references · 8 canonical work pages

  1. [1]

    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. doi:10.1...

    work page doi:10.1145/3292500.3330701 2019

  2. [2]

    Alessio Benavoli, Giorgio Corani, Janez Demšar, and Marco Zaffalon. 2017. Time for a Change: a Tutorial for Comparing Multiple Classifiers Through Bayesian Analysis.Journal of Machine Learning Research18, 77 (2017), 1–36. http://jmlr. org/papers/v18/16-305.html

    2017

  3. [3]

    Ralph Allan Bradley and Milton E Terry. 1952. Rank analysis of incomplete block designs: I. the method of paired comparisons.Biometrika39, 3/4 (1952), 324–345

    1952

  4. [4]

    Francois Caron and Arnaud Doucet. 2012. Efficient Bayesian inference for gener- alized Bradley–Terry models.Journal of Computational and Graphical Statistics 21, 1 (2012), 174–196. KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Grishina et al

    2012

  5. [5]

    Giuseppe Casalicchio, Gerhard Tutz, and Gunther Schauberger. 2015. Subject- specific Bradley–Terry–Luce models with implicit variable selection.Statistical Modelling15, 6 (2015), 526–547

    2015

  6. [6]

    Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Hao Zhang, Banghua Zhu, Michael Jordan, Joseph E Gonzalez, et al. 2024. Chatbot arena: An open platform for evaluating llms by human preference.arXiv preprint arXiv:2403.04132(2024)

    Pith/arXiv arXiv 2024

  7. [7]

    Paolo Cremonesi, Yehuda Koren, and Roberto Turrin. 2010. Performance of recommender algorithms on top-n recommendation tasks. InProceedings of the Fourth ACM Conference on Recommender Systems(Barcelona, Spain)(RecSys ’10). Association for Computing Machinery, New York, NY, USA, 39–46. doi:10.1145/ 1864708.1864721

    arXiv 2010

  8. [8]

    Roger R Davidson. 1970. On extending the Bradley-Terry model to accommodate ties in paired comparison experiments.J. Amer. Statist. Assoc.65, 329 (1970), 317–328

    1970

  9. [9]

    Gerard Debreu. 1960. Individual choice behavior: A theoretical analysis

    1960

  10. [10]

    Janez Demšar. 2006. Statistical Comparisons of Classifiers over Multiple Data Sets.Journal of Machine Learning Research7, 1 (2006), 1–30. http://jmlr.org/ papers/v7/demsar06a.html

    2006

  11. [11]

    Jianqing Fan, Jikai Hou, and Mengxin Yu. 2024. Uncertainty quantification of MLE for entity ranking with covariates.Journal of Machine Learning Research 25, 358 (2024), 1–83

    2024

  12. [12]

    Evgeny Frolov and Ivan Oseledets. 2023. Tensor-Based Sequential Learning via Hankel Matrix Representation for Next Item Recommendations.IEEE Access11 (2023), 6357–6371. doi:10.1109/ACCESS.2023.3234863

    work page doi:10.1109/access.2023.3234863 2023

  13. [13]

    Chao Gao, Yandi Shen, and Anderson Y Zhang. 2023. Uncertainty quantification in the Bradley–Terry–Luce model.Information and Inference: A Journal of the IMA12, 2 (2023), 1073–1140

    2023

  14. [14]

    Pieter Gijsbers, Marcos LP Bueno, Stefan Coors, Erin LeDell, Sébastien Poirier, Janek Thomas, Bernd Bischl, and Joaquin Vanschoren. 2024. Amlb: an automl benchmark.Journal of Machine Learning Research25, 101 (2024), 1–65

    2024

  15. [15]

    Mark E Glickman. [n. d.]. Paired comparison models with strength-dependent ties and order effects.Statistical Modelling([n. d.]), 1471082X251400474

  16. [16]

    Danil Gusak, Anna Volodkevich, Anton Klenitskiy, Alexey Vasilev, and Evgeny Frolov. 2025. Time to Split: Exploring Data Splitting Strategies for Offline Evalu- ation of Sequential Recommenders. InProceedings of the Nineteenth ACM Confer- ence on Recommender Systems (RecSys ’25). Association for Computing Machinery, New York, NY, USA, 874–883. doi:10.1145/...

    work page doi:10.1145/3705328.3748164 2025

  17. [17]

    Malay Haldar, Daochen Zha, Huiji Gao, Liwei He, and Sanjeev Katariya. 2025. Beyond Pairwise Learning-To-Rank At Airbnb. InProceedings of the 34th ACM International Conference on Information and Knowledge Management(Seoul, Re- public of Korea)(CIKM ’25). ACM, New York, NY, USA, 8 pages. doi:10.1145/ 3746252.3761521

    arXiv 2025

  18. [18]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, YongDong Zhang, and Meng Wang. 2020. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation(SIGIR ’20). Association for Computing Machinery, New York, NY, USA, 639–648. doi:10.1145/3397271.3401063

    work page doi:10.1145/3397271.3401063 2020

  19. [19]

    Reinhard Heckel, Max Simchowitz, Kannan Ramchandran, and Martin Wain- wright. 2018. Approximate Ranking from Pairwise Comparisons. InProceedings of the Twenty-First International Conference on Artificial Intelligence and Sta- tistics (Proceedings of Machine Learning Research, Vol. 84), Amos Storkey and Fernando Perez-Cruz (Eds.). PMLR, 1057–1066. https://...

    2018

  20. [20]

    Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative filtering for implicit feedback datasets. In2008 Eighth IEEE international conference on data mining. Ieee, 263–272

    2008

  21. [21]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Recom- mendation. InProceedings of the IEEE International Conference on Data Mining (ICDM). IEEE, 197–206

    2018

  22. [22]

    Anton Klenitskiy, Anna Volodkevich, Anton Pembek, and Alexey Vasilev. 2024. Does it look sequential? an analysis of datasets for evaluation of sequential recommendations. InProceedings of the 18th ACM Conference on Recommender Systems. 1067–1072

    2024

  23. [23]

    PS Kostenetskiy, RA Chulkevich, and VI Kozyrev. 2021. HPC resources of the higher school of economics. InJournal of Physics: Conference Series, Vol. 1740. IOP Publishing, 012050

    2021

  24. [24]

    Kelong Mao, Jieming Zhu, Xi Xiao, Biao Lu, Zhaowei Wang, and Xiuqiang He

  25. [25]

    InProceedings of the 30th ACM international conference on information & knowledge management

    UltraGCN: ultra simplification of graph convolutional networks for recom- mendation. InProceedings of the 30th ACM international conference on information & knowledge management. 1253–1262

  26. [26]

    Oseledets, and Evgeny Frolov

    Gleb Mezentsev, Danil Gusak, Ivan V. Oseledets, and Evgeny Frolov. 2024. Scalable Cross-Entropy Loss for Sequential Recommendations with Large Item Catalogs. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys). ACM, 475–485

    2024

  27. [27]

    Frederick Mosteller. 1951. Remarks on the Method of Paired Comparisons: I. The Least Squares Solution Assuming Equal Standard Deviations and Equal Correlations.Psychometrika16, 1 (1951), 3–9

    1951

  28. [28]

    Sewoong Oh. 2017. Rank centrality: Ranking from pairwise comparisons.Opera- tions research(2017)

    2017

  29. [29]

    Robin L Plackett. 1975. The analysis of permutations.Journal of the Royal Statistical Society Series C: Applied Statistics24, 2 (1975), 193–202

    1975

  30. [30]

    Wang Qinsi, Jinghan Ke, Masayoshi Tomizuka, Kurt Keutzer, and Chenfeng Xu

  31. [31]

    InThe Thirteenth International Conference on Learning Representations

    Dobi-SVD: Differentiable SVD for LLM Compression and Some New Per- spectives. InThe Thirteenth International Conference on Learning Representations. https://openreview.net/forum?id=kws76i5XB8

  32. [32]

    P. V. Rao and Lawrence L. Kupper. 1967. Ties in Paired-Comparison Experiments: A Generalization of the Bradley–Terry Model.J. Amer. Statist. Assoc.62, 317 (1967), 194–204

    1967

  33. [33]

    Steffen Rendle and Christoph Freudenthaler. 2014. Improving pairwise learning for item recommendation from implicit feedback. InProceedings of the 7th ACM International Conference on Web Search and Data Mining(New York, New York, USA)(WSDM ’14). Association for Computing Machinery, New York, NY, USA, 273–282. doi:10.1145/2556195.2556248

    work page doi:10.1145/2556195.2556248 2014

  34. [34]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme

  35. [35]

    InProceed- ings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence (UAI)

    BPR: Bayesian Personalized Ranking from Implicit Feedback. InProceed- ings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence (UAI). Montreal, QC, Canada, 452–461

  36. [36]

    Gunther Schauberger and Gerhard Tutz. 2019. BTLLasso: a common framework and software package for the inclusion and selection of covariates in Bradley- Terry models.Journal of Statistical Software88 (2019), 1–29

    2019

  37. [37]

    Mete Sertkan, Sophia Althammer, Sebastian Hofstätter, Peter Knees, and Julia Neidhardt. 2023. Exploring Effect-Size-Based Meta-Analysis for Multi-Dataset Evaluation. InProceedings of the 3rd Workshop Perspectives on the Evaluation of Rec- ommender Systems 2023 co-located with the 17th ACM Conference on Recommender Systems (RecSys 2023) (CEUR Workshop Proc...

    2023

  38. [38]

    Valeriy Shevchenko, Nikita Belousov, Alexey Vasilev, Vladimir Zholobov, Artyom Sosedka, Natalia Semenova, Anna Volodkevich, Andrey Savchenko, and Alexey Zaytsev. 2024. From Variability to Stability: Advancing RecSys Benchmarking Practices. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Dis- covery and Data Mining(Barcelona, Spain)(KDD ’24). ...

    work page doi:10.1145/3637528.3671655 2024

  39. [39]

    Vladimir Spokoiny. 2025. Semiparametric plug-in estimation, sup-norm risk bounds, marginal optimization, and inference in BTL model.arXiv preprint arXiv:2503.15045(2025)

    arXiv 2025

  40. [40]

    Harald Steck. 2019. Embarrassingly Shallow Autoencoders for Sparse Data. InThe World Wide Web Conference(San Francisco, CA, USA)(WWW ’19). Association for Computing Machinery, New York, NY, USA, 3251–3257. doi:10.1145/3308558. 3313710

    work page doi:10.1145/3308558 2019

  41. [41]

    Carolin Strobl, Florian Wickelmaier, and Achim Zeileis. 2011. Accounting for individual differences in Bradley-Terry models by means of recursive partitioning. Journal of Educational and Behavioral Statistics36, 2 (2011), 135–153

    2011

  42. [42]

    Gábor Takács, István Pilászy, and Domonkos Tikk. 2011. Applications of the con- jugate gradient method for implicit feedback collaborative filtering. InProceedings of the fifth ACM conference on Recommender systems. 297–300

    2011

  43. [43]

    Thurstone

    Louis L. Thurstone. 1927. A Law of Comparative Judgment.Psychological Review 34 (1927), 273–286

    1927

  44. [44]

    Askar Tsyganov, Evgeny Frolov, Sergey Samsonov, and Maxim Rakhuba. 2026. Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation.Proceedings of the AAAI Conference on Artificial Intelligence40, 31 (2026), 26010–26018

    2026

  45. [45]

    Askar Tsyganov, Uliana Parkina, Ekaterina Grishina, Sergey Samsonov, and Maxim Rakhuba. 2026. Faster SVD via Accelerated Newton-Schulz Iteration. In ICLR Blogposts 2026(April 27, 2026). https://iclr-blogposts.github.io/2026/blog/ 2026/polar-svd/

    2026

  46. [46]

    Tobias Vente, Michael Heep, Abdullah Abbas, Theodor Sperle, Joeran Beel, and Bart Goethals. 2025. APS Explorer: Navigating Algorithm Performance Spaces for Informed Dataset Selection. InProceedings of the Nineteenth ACM Conference on Recommender Systems. 1322–1324

    2025

  47. [47]

    Jacques Wainer. 2023. A Bayesian Bradley–Terry Model to Compare Multiple Machine Learning Algorithms on Multiple Data Sets.Journal of Machine Learning Research24 (10 2023), 1–34

    2023

  48. [48]

    Jordan, and Nebojsa Jojic

    Fabian Wauthier, Michael I. Jordan, and Nebojsa Jojic. 2013. Efficient Ranking from Pairwise Comparisons. InProceedings of the 30th International Conference on Machine Learning (ICML) (Proceedings of Machine Learning Research, Vol. 28). 109–117

    2013

  49. [49]

    Achim Zeileis, Torsten Hothorn, and Kurt Hornik. 2008. Model-based Recursive Partitioning.Journal of Computational and Graphical Statistics17, 2 (2008), 492–514. doi:10.1198/10618600SX319331

    work page doi:10.1198/10618600sx319331 2008

  50. [50]

    A Zeileis, C Strobl, F Wickelmaier, and J Kopf. 2011. psychotree: Recursive parti- tioning based on psychometric models.R package version 0.12-1, URL http://CRAN. R-project. org/package= psychotree(2011)

    2011

  51. [51]

    Implicit

    Ernst Zermelo. 1929. Die berechnung der turnier-ergebnisse als ein maxi- mumproblem der wahrscheinlichkeitsrechnung.Mathematische Zeitschrift29, 1 (1929), 436–460. Bradley-Terry Rankings for Recommender Systems Across Dataset Taxonomies KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea A Supplementary tables Table 7: Recommendation algorithms us...

    arXiv 1929