Pith. sign in

REVIEW 3 major objections 4 minor 28 references

Adaptive Experiments Under Data Sparse Settings: Applications for Educational Platforms

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read WAPTS, a reweighted Thompson Sampling rule, identifies near-optimal treatments faster and more reliably than standard Thompson Sampling in data-sparse educational experiments, according to simulation studies.

desk verdict A simple reweighting of Thompson Sampling that looks practical in sparse-data settings, but its own experiments contradict the 'consistently outperforms' claim, and the zero-weight rule can permanently kill the true best arm. read the letter →

arxiv 2501.03999 v4 pith:UPGNECFU submitted 2025-01-07 cs.LG stat.ML

Haochen Song , Ilya Musabirov , Ananya Bhattacharjee , Audrey Durand , Meredith Franklin , Anna Rafferty , Joseph Jay Williams This is my paper · ORCID
classification cs.LGstat.ML
keywords Thompsonsamplingmulti-armedbanditsadaptiveexperimentsdata-sparsesettingseducationalplatformslearnersourcinglenientregretposteriorweighting
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

The paper argues that standard Thompson Sampling wastes scarce student evaluations in educational experiments with many treatment arms and few participants, and proposes WAPTS (Weighted Allocation Probability Adjusted Thompson Sampling) to fix that. WAPTS adds a deterministic weight based on each arm's empirical success rate to the usual Thompson draw, pulling allocation toward treatments that are already looking good while still sampling from posteriors. Across simulation studies, the authors find WAPTS selects a near-optimal treatment with higher probability, converges its posterior faster, and keeps hypothesis-testing power comparable to Thompson Sampling. The motivation is practical: if true, instructors running small adaptive experiments could identify a small set of 'good enough' learning materials without needing large samples.

What carries the argument

The central object is the weighted allocation score $\omega_i = (1 + r_i) r_i \tilde p_i$, defined for each treatment $i$. Here $\tilde p_i$ is drawn from the Beta posterior $\mathrm{Beta}(k_s[i]+1, k_f[i]+1)$, and $r_i = k_s[i]/(k_s[i]+k_f[i])$ is the empirical success rate. Multiplying the Thompson sample by an increasing function of the observed success rate reorders the draws: an arm with 80% observed success gets a larger score than one at 20% even when their posterior samples are equal, so allocation shifts toward empirically strong treatments. This is the mechanism that carries the paper's claim of faster convergence; it is also the mechanism that can in rare cases drive a truly optimal treatment's score to zero if early outcomes are negative.

What would settle it

Simulate a fixed-gap bandit with $K=10$ arms, $N=239$ participants, and effect size $\Delta=0.1$ across many seeds, with the optimal arm forced to receive three failures before any success in early rounds; measure the fraction of runs in which WAPTS reduces the optimal arm's allocation to zero for the rest of the horizon. If that fraction is not negligible, the claim that WAPTS reliably identifies near-optimal treatments fails for unlucky early outcomes.

Watch

Extended reading notes

Core claim

In the paper's own terms, WAPTS is a bandit policy that outperforms traditional Thompson Sampling in data-sparse educational settings on three metrics: higher probability of correctly selecting a near-optimal treatment, faster posterior convergence to near-optimal treatments, and comparable stability in hypothesis testing. The mechanism is a weighted allocation score $\omega_i = (1 + r_i) r_i \tilde p_i$, where $r_i$ is the empirical success rate and $\tilde p_i$ is a Thompson sample from the Beta posterior; after scoring all arms, the policy plays the arm with the largest $\omega_i$. Because the weight grows with empirical success, poor treatments are discarded earlier and promising treatments are exploited sooner, trading asymptotic optimality for faster practical discovery. The authors emphasize this is intended for 'lenient regret' settings where several treatments within an $\epsilon$-band of the best are acceptable, which they argue matches how educators actually choose content.

Load-bearing premise

The algorithm assumes that a treatment's observed success rate over the first few students is informative enough to reweight by, so that downweighting an unlucky arm toward zero cannot permanently hide the truly best treatment.

Editorial extensions

If this is right

  • In learnersourcing tasks where students rate peer-generated examples, WAPTS can surface a small set of useful materials after fewer ratings than Thompson Sampling, under the simulation conditions reported.
  • Practitioners can set the lenient tolerance $\epsilon$ from sample size, number of treatments, and effect-size estimates rather than choosing it arbitrarily.
  • For data-sparse deployments (e.g., $N=50$ with $K=2$, $N=300$ with $K=50$), WAPTS shows higher correct-selection rates than uniform random assignment and usually than Thompson Sampling, though absolute correct-selection rates remain low when $K/N$ is large.
  • The theoretical $O(\log n)$-type regret bound requires an additional safeguard like forced exploration or allocation resets; without it, WAPTS can in principle suffer unbounded regret in a rare unlucky run.

Reading between the lines

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

  • Editorial extension: the same outcome-sensitive reweighting idea could be applied to contextual bandits or continuous outcomes, as the authors note, but the binary-success weight would need replacing with an effect-size-aware penalty.
  • Editorial extension: a reset mechanism of the kind proposed could be implemented cost-free and tested in the same simulations; if reset thresholds are chosen well, it may recover the rare-case failures without losing the early-exploitation gains.
  • Editorial extension: the paper's simulations compare WAPTS with Thompson Sampling and uniform random assignment only; a direct comparison with UCB-style or other probability-matching rules under the same data-sparse settings would sharpen the claim that the weighting, not just adaptivity, drives the improvement.
  • Editorial extension: because the weight is deterministic and monotone, WAPTS effectively turns Thompson Sampling into a more exploitative rule; one testable prediction is that its advantage grows as the effect size $\Delta$ grows at small-to-moderate values, but reverses for very large $\Delta$ where Thompson Sampling's exploration pays off.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes Weighted Allocation Probability Adjusted Thompson Sampling (WAPTS), a variant of Thompson Sampling in which each arm's posterior draw is multiplied by an empirical-success weight (1+r_i)r_i, with the goal of improving identification of near-optimal treatments under data-sparse educational experiments. The authors evaluate WAPTS against Thompson Sampling and uniform random allocation through simulations in learnersourcing and related settings, claiming faster convergence, higher probability of selecting near-optimal treatments, and comparable hypothesis-testing stability. The manuscript also provides a lenient-regret framework and a theoretical regret sketch in the appendix.

Significance. If the central claims were fully supported, WAPTS would be a simple, practical heuristic for adaptive experiments with many treatments and small samples, which is a genuinely relevant problem in educational technology. The paper's strength is its clear focus on lenient regret and on identifying sets of good-enough treatments rather than a single winner, and it includes extensive simulation tables across several application-inspired configurations. However, the evidence presented is internally mixed: several reported results contradict the abstract's 'consistently outperforms' claim, and the algorithm's zero-weight rule can permanently discard the truly optimal arm after early bad luck. These issues directly undermine the paper's main conclusions as currently stated.

major comments (3)
  1. The WAPTS selection score is defined as ω_i = (1 + r_i) r_i \tilde{p}_i, where r_i = ks[i]/(ks[i]+kf[i]). If an arm has at least one failure and zero successes, then r_i = 0 and the weight is exactly 0, so the arm can never be selected again via arg max. Because any Bernoulli arm with θ_i < 1 has positive probability of an initial failure, the truly optimal arm can be permanently discarded with positive probability. The paper itself acknowledges this in the Discussion ('W APTS currently updates in a direction that is difficult to reverse... could risk prematurely discarding promising treatments') and in Appendix A.4, yet Algorithm 1 contains no safeguard such as forced exploration, a reset mechanism, or an additive epsilon. This structural property invalidates the claims of 'higher probability of correctly selecting near-optimal treatment' and 'faster posterior convergence to near-optimal treatments' on exactly the unlucky seeds where a good treatment fails early, and it also means the empirical averages over replications are not representative of reliable identification.
  2. The abstract states that 'WAPTS consistently outperforms traditional TS' on key metrics, but the paper's own results contradict this. In Table 1, at effect sizes 0.20 and 0.30, TS has higher power than WAPTS for detecting at least one ϵ-lenient treatment (39.6% vs 38.1% and 78.3% vs 74.1%). In Appendix F, TS matches or exceeds WAPTS on mean outcome in several configurations: Table 6 (TS 0.617 vs WAPTS 0.611), Table 8 (both 0.565), and Table 9 (both 0.512). The text in the Power Trends section even states that 'at larger effects, TS more often identifies the true best treatment,' which is opposite to Table 1's true-best column (WAPTS 24.8% vs TS 16.5% at ∆=0.20, and 57.5% vs 43.6% at ∆=0.30). This internal inconsistency means the paper's headline claim of consistent superiority is not supported by the reported evidence; the authors need to either revise the claim to a more conditional statement or re-analyze the data.
  3. The regret analysis assumes a burn-in phase of b rounds with uniform sampling ('For the first b rounds, W APTS selects treatments uniformly'), but Algorithm 1 has no such burn-in. The O(log n) regret bound is therefore derived for a different policy than the one implemented and evaluated. Additionally, the proof sketch relies on concentration inequalities applied to the reweighted scores w_k = (1 + r_k) r_k \tilde{p}_k without providing a justification that these inequalities hold uniformly for the data-dependent weighting; the sketch states that 'using standard concentration inequalities' the mistake probability is bounded, but the reweighting factor itself depends on the same empirical counts and can be zero for the optimal arm. Thus the theoretical support for the implemented algorithm is not established.
minor comments (4)
  1. There are repeated typos and inconsistent formatting: 'policys' should be 'policies', 'treatmentes' in Appendix A.1, and the policy name is written as both 'W APTS' and 'WAPTS'; these should be unified.
  2. The sentence 'at larger effects, TS more often identifies the true best treatment' is inconsistent with Table 1, where WAPTS has higher power for the true best treatment at ∆=0.20 and ∆=0.30; please correct the text or the table.
  3. The caption refers to a 'vertical dashed line marks burn-in,' but Algorithm 1 has no burn-in phase. Please clarify whether the simulations included a burn-in period and, if so, why it is not in the pseudocode.
  4. The correct-assignment rates are extremely low (around 2–5%) for all policies in the K=50 learnersourcing setting, and the mean-outcome differences between policies are very small; the text should discuss whether the reported 'superior counts' are practically meaningful given these near-chance rates.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: WAPTS claims are evaluated against known simulation ground truth and do not reduce to fitted inputs or self-citations.

full rationale

The paper's central claims about WAPTS are supported by simulation comparisons against Thompson Sampling and uniform random allocation under known Bernoulli ground-truth parameters. The weighting function f(p) = (1+p)p is the algorithm's mechanism, not a parameter fitted to the evaluation data; the reported metrics—average outcome, correct-selection rate, posterior evolution, and lenient regret—are measured outcomes rather than quantities defined in terms of the weight itself. Even the "faster convergence" result is a contingent empirical consequence of allocating more samples to empirically strong arms, not an identity: the paper's own figures and tables show settings where WAPTS does not dominate (e.g., larger effect sizes where TS sometimes has lower regret or higher commitment rates). The theoretical sketch in Appendix A adapts the external Thompson Sampling analysis of Agrawal and Goyal (2012) and includes a burn-in phase that is absent from Algorithm 1; that is a proof-pseudocode mismatch and a correctness risk, not circularity. The paper explicitly acknowledges the zero-weight failure mode in the Discussion and Appendix A.4, which undermines the reliability claim but does not make the derivation circular. Self-citations such as Rafferty, Ying, and Williams (2019) and Williams et al. (2021) motivate the problem setting but are not the load-bearing evidence for WAPTS's reported advantages, which are demonstrated in the paper's own externally benchmarked simulations. No prediction reduces by construction to an input, and no load-bearing argument rests on an unverified self-citation chain.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the chosen reweighting function and on the simulation parameters (eta, Delta, N, K). The only invented entity is the algorithm itself, which is a reweighting of Thompson Sampling. No new physical or mathematical entities are introduced.

free parameters (3)
  • lenient tolerance fraction eta = 0.5
    Set as eta = 0.5 in the simulations to define epsilon = eta * Delta. Chosen by hand, not derived from any principle; the paper states leniency 'can be derived' but provides no formula.
  • burn-in length b in regret proof = unspecified
    Appendix A.3 assumes a uniform burn-in phase for the first b rounds, but Algorithm 1 has no burn-in. The value b is never specified, and the proof only works if this extra mechanism is added.
  • reweighting function f(r) = (1 + r) * r
    Ad hoc choice of the weighting function. No derivation or comparison to alternative weights is provided, and the function drives all observed performance differences.
assumptions (3)
  • domain assumption Rewards are i.i.d. Bernoulli with a fixed success probability per treatment.
    Stated in the simulation setup and used throughout. Real educational outcomes may be nonstationary or correlated across students.
  • domain assumption Treatments are independent with no shared structure or covariates.
    The paper focuses on non-contextual bandits and explicitly sets aside contextual information. This limits applicability to many educational settings.
  • ad hoc to paper Concentration inequalities apply to the reweighted scores w_k = (1 + r_k) * r_k * p_k.
    Step 3 of Appendix A.3 asserts that after enough samples both p_k and theta_k concentrate near p_k, and the increasing monotonicity of f widens the gap. This is asserted, not proved, and the zero-weight case violates the assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Experiments Under Data Sparse Settings: Applications for Educational Platforms." pith.science (2026). https://pith.science/paper/UPGNECFU

@misc{pith2026250103999,
  author       = {Pith},
  title        = {Pith review of: Adaptive Experiments Under Data Sparse Settings: Applications for Educational Platforms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UPGNECFU}},
  note         = {Machine review of arXiv:2501.03999}
}
read the original abstract

Adaptive experimentation is increasingly used in educational platforms to personalize learning through dynamic content and feedback. However, standard adaptive strategies such as Thompson Sampling often underperform in real-world educational settings where content variations are numerous and student participation is limited, resulting in sparse data. In particular, Thompson Sampling can lead to imbalanced content allocation and delayed convergence on which aspects of content are most effective for student learning. To address these challenges, we introduce Weighted Allocation Probability Adjusted Thompson Sampling (WAPTS), an algorithm that refines the sampling strategy to improve content-related decision-making in data-sparse environments. WAPTS is guided by the principle of lenient regret, allowing near-optimal allocations to accelerate learning while still exploring promising content. We evaluate WAPTS in a learnersourcing scenario where students rate peer-generated learning materials, and demonstrate that it enables earlier and more reliable identification of promising treatments.

Figures

Figures reproduced from arXiv: 2501.03999 by the authors.

Figure 1
Figure 1. Average outcome vs. Number of Treatments (sam [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Violin plot of participant allocations in sub [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Policy comparison across varying sample sizes [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Learnersourcing example flow: N students cre￾ate constructive retrieval examples. After an initial screening process, K examples are selected and then passed forward for peer ratings. We define a family of hypothesis tests, one per treatment: H0,i : θi ≤ θ ∗ − ϵ H1,i :…
Figure 7
Figure 7. Figure 7: summarizes posterior evolution for the flexible gap setting, under N = 239 with four effect sizes (∆ ∈ {0.05, 0.1, 0.2, 0.3}). In both scenarios: • WAPTS shows faster convergence and lower variance. • TS remains unstable or conservative, especially at lower effect size…
Figure 8
Figure 8. Figure 8: Cumulative ϵ-lenient regret over time under the flexible-gap setting (K = 10, N = 239, ϵ = 0.5 × ∆). Lines show Monte Carlo means for TS (blue) and WAPTS (red); ribbons are 95% bands; the vertical dashed line indi￾cates burn-in. We also look at the convergence speed of…
Figure 9
Figure 9. Figure 9: Posterior evolution (Fixed-Gap) with N = 239. WAPTS shows tighter estimates and earlier convergence. E Appendix: Majority-Commit Metric and Lenient Correctness Majority-commit time. To evaluate the convergence speed and compare WAPTS with TS, we define the follow￾ing m…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 24 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    W.; Barnes, T.; and Chi, M

    Abdelshiheed, M.; Hostetter, J. W.; Barnes, T.; and Chi, M. 2023. Leveraging Deep Reinforcement Learning for Metacognitive Interventions Across Intelligent Tutoring Systems . In Wang, N.; Rebolledo-Mendez, G.; Matsuda, N.; Santos, O. C.; and Dimitrova, V., eds., Artificial Intelligence in Education , 291--303. Cham: Springer Nature Switzerland. ISBN 978-3...

    work page 2023

  4. [4]

    Agrawal, S.; and Goyal, N. 2012. Analysis of Thompson Sampling for the multi-armed bandit problem. ArXiv:1111.1797 [cs]

    arXiv 2012

  5. [5]

    Chapelle, O.; and Li, L. 2011. An Empirical Evaluation of Thompson Sampling . In Shawe-Taylor, J.; Zemel, R.; Bartlett, P.; Pereira, F.; and Weinberger, K. Q., eds., Advances in Neural Information Processing Systems , volume 24. Curran Associates, Inc

    work page 2011

  6. [6]

    Dimakopoulou, M.; Ren, Z.; and Zhou, Z. 2021. Online Multi - Armed Bandits with Adaptive Inference . In Advances in Neural Information Processing Systems , volume 34, 1939--1951. Curran Associates, Inc

    work page 2021

  7. [7]

    Doroudi, S.; Kamar, E.; and Brunskill, E. 2019. Not Everyone Writes Good Examples but Good Examples Can Come from Anywhere . Proceedings of the AAAI Conference on Human Computation and Crowdsourcing, 7: 12--21

    work page 2019

  8. [8]

    Endres, T.; Carpenter, S.; and Renkl, A. 2024. Constructive retrieval: Benefits for learning, motivation, and metacognitive monitoring. Learning and Instruction, 94: 101974

    work page 2024

Show all 28 references
  1. [9]

    Erraqabi, A.; Lazaric, A.; Valko, M.; Brunskill, E.; and Liu, Y.-E. 2017. Trading off Rewards and Errors in Multi - Armed Bandits . In Proceedings of the 20th International Conference on Artificial Intelligence and Statistics , 709--717. PMLR. ISSN: 2640-3498

    2017

  2. [10]

    Feng, Q.; Ma, T.; and Zhu, R. 2025. Satisficing Regret Minimization in Bandits : Constant Rate and Light - Tailed Distribution . ArXiv:2406.06802 [stat]

    2025 arXiv

  3. [11]

    Fiorella, L.; and Mayer, R. E. 2016. Eight Ways to Promote Generative Learning . Educational Psychology Review, 28(4): 717--741

    2016

  4. [12]

    Kaufmann, E.; Cappé, O.; and Garivier, A. 2016. On the Complexity of Best Arm Identification in Multi - Armed Bandit Models . ArXiv:1407.4443 [stat]

    2016 arXiv

  5. [13]

    Khosravi, H.; Denny, P.; Moore, S.; and Stamper, J. 2023. Learnersourcing in the age of AI: Student, educator and machine partnerships for content creation. Computers and Education: Artificial Intelligence, 5: 100151

    2023

  6. [14]

    Kumar, H.; Li, T.; Shi, J.; Musabirov, I.; Kornfield, R.; Meyerhoff, J.; Bhattacharjee, A.; Karr, C.; Nguyen, T.; Mohr, D.; Rafferty, A.; Villar, S.; Deliu, N.; and Williams, J. J. 2024. Using Adaptive Bandit Experiments to Increase and Investigate Engagement in Mental Health ...

    2024

  7. [15]

    Lattimore, T.; and Szepesvári, C. 2020. Bandit Algorithms . Cambridge University Press, 1 edition. ISBN 978-1-108-57140-1 978-1-108-48682-8

    2020

  8. [16]

    Li, L.; Chu, W.; Langford, J.; and Schapire, R. E. 2010. A Contextual - Bandit Approach to Personalized News Article Recommendation . In Proceedings of the 19th international conference on World wide web , 661--670. ArXiv:1003.0146 [cs]

    2010 arXiv

  9. [17]

    Mandel, T.; Liu, Y.-E.; Brunskill, E.; and Popović, Z. 2017. Where to Add Actions in Human -in-the- Loop Reinforcement Learning . Proceedings of the AAAI Conference on Artificial Intelligence, 31(1)

    2017

  10. [18]

    Merlis, N.; and Mannor, S. 2021. Lenient Regret for Multi - Armed Bandits . ArXiv:2008.03959 [cs]

    2021 arXiv

  11. [19]

    Palmisano, C.; Tuzhilin, A.; and Gorgoglione, M. 2008. Using Context to Improve Predictive Modeling of Customers in Personalization Applications. IEEE Transactions on Knowledge and Data Engineering, 20(11): 1535--1549

    2008

  12. [20]

    Rafferty, A.; Ying, H.; and Williams, J. 2019. Statistical Consequences of using Multi -armed Bandits to Conduct Adaptive Educational Experiments . Journal of Educational Data Mining, 11(1): 47--79

    2019

  13. [21]

    N.; and Williams, J

    Reza, M.; Kim, J.; Bhattacharjee, A.; Rafferty, A. N.; and Williams, J. J. 2021. The MOOClet Framework : Unifying Experimentation , Dynamic Improvement , and Personalization in Online Courses . In Proceedings of the Eighth ACM Conference on Learning @ Scale , 15--26. Virtual E...

    2021

  14. [22]

    Y.; Ying, R.; et al

    Ruan, S.; Nie, A.; Steenbergen, W.; He, J.; Zhang, J.; Guo, M.; Liu, Y.; Dang Nguyen, K.; Wang, C. Y.; Ying, R.; et al. 2024. Reinforcement learning tutor better supported lower performers in a math task. Machine Learning, 113(5): 3023--3048

    2024

  15. [23]

    Shawky, D.; and Badawi, A. 2019. Towards a Personalized Learning Experience Using Reinforcement Learning . In Hassanien, A. E., ed., Machine Learning Paradigms : Theory and Application , 169--187. Cham: Springer International Publishing. ISBN 978-3-030-02357-7

    2019

  16. [24]

    D.; Frau, A.; Burns, N.; Faraji, F.; Potochny, J.; Robert, S.; Roussel, A.; Zheng, S.; Glazier, T.; Romano, J

    St-Hilaire, F.; Vu, D. D.; Frau, A.; Burns, N.; Faraji, F.; Potochny, J.; Robert, S.; Roussel, A.; Zheng, S.; Glazier, T.; Romano, J. V.; Belfer, R.; Shayan, M.; Smofsky, A.; Delarosbil, T.; Ahn, S.; Eden-Walker, S.; Sony, K.; Ching, A. O.; Elkins, S.; Stepanyan, A.; Matajova,...

    2022 arXiv

  17. [25]

    Tang, L.; Jiang, Y.; Li, L.; and Li, T. 2014. Ensemble contextual bandits for personalized recommendation. In Proceedings of the 8th ACM Conference on Recommender systems , RecSys '14, 73--80. New York, NY, USA: Association for Computing Machinery. ISBN 978-1-4503-2668-1

    2014

  18. [26]

    J.; Kim, J.; Rafferty, A.; Maldonado, S.; Gajos, K

    Williams, J. J.; Kim, J.; Rafferty, A.; Maldonado, S.; Gajos, K. Z.; Lasecki, W. S.; and Heffernan, N. 2016. AXIS: Generating explanations at scale with learnersourcing and machine learning. In Proceedings of the third (2016) ACM conference on learning@ scale, 379--388

    2016

  19. [27]

    J.; Nogas, J.; Deliu, N.; Shaikh, H.; Villar, S

    Williams, J. J.; Nogas, J.; Deliu, N.; Shaikh, H.; Villar, S. S.; Durand, A.; and Rafferty, A. 2021. Challenges in Statistical Analysis of Data Collected by a Bandit Algorithm : An Empirical Exploration in Applications to Adaptively Randomized Experiments . ArXiv:2103.12198 [cs, stat]

    2021 arXiv

  20. [28]

    Zhao, C.; Zhao, H.; He, M.; Zhang, J.; and Fan, J. 2023. Cross-domain recommendation via user interest alignment. In Proceedings of the ACM Web Conference 2023 , 887--896. Austin TX USA: ACM. ISBN 978-1-4503-9416-1

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.