Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Impatient Bandits: Optimizing for the Long-Term Without Delay

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A Thompson-sampling Bayesian filter learns 60-day rewards from early engagement; a large podcast A/B test saw recent-show 60-day metrics rise over 50%.

desk verdict Formal progressive-feedback bandits with a genuinely novel VoPF-dependent regret bound and a production-scale A/B test; the main soft spot is the correctly-specified-prior requirement, which the authors openly acknowledge. read the letter →

arxiv 2501.07761 v2 pith:STAOABRL submitted 2025-01-14 cs.LG cs.AIstat.ML

Kelly W. Zhang , Thomas Baldwin-McDonald , Kamil Ciosek , Lucas Maystre , Daniel Russo This is my paper · ORCID
classification cs.LGcs.AIstat.ML MSC 68Q32
keywords multi-armedbanditsdelayedrewardsprogressivefeedbackThompsonsamplingBayesianfilteringregretboundscold-startrecommendationrecommendersystems
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

This paper takes on the cold-start problem in recommender systems, where the reward of interest—say, engagement over 60 days—arrives long after the recommendation decision. The authors model the setting as a bandit with progressive feedback: short-term engagement outcomes arrive day by day and are imperfectly correlated with the eventual long-term reward. Their algorithm, the Impatient Bandit, combines Thompson sampling with a Bayesian filter that updates a posterior over each item's latent quality as partial trajectories arrive. The paper's central theoretical claim is that the expected regret in every batch is at most the regret of delayed-reward Thompson sampling scaled by $\exp(-\mathrm{VoPF}(t))$, where $\mathrm{VoPF}(t)$ is the mutual information between progressive feedback and the mean reward. If correct, this means informative early signals can drive good decisions well before any long-term reward is ever observed, and the reported podcast A/B test supports that conclusion with over 50% improvements in 60-day engagement metrics for recently released shows.

What carries the argument

The load-bearing object is the Value of Progressive Feedback, $\mathrm{VoPF}(t) = I(\bar R_a; \{\tilde Y_{u,t}(a)\}_{u\in\mathcal{U}_{<t}} \mid \{\tilde Y_{u,t}(a)\}_{u\in\mathcal{U}_{<(t-d_{\max})}})$, an environment-defined conditional mutual information that measures how much of the uncertainty in an item's mean long-term reward is resolved by early engagement beyond what the fully delayed rewards would resolve. It enters the bound directly as $\exp(-\mathrm{VoPF}(t))$. The second engine is the Bayesian filter: a Cholesky whitening of the noise covariance converts each partial engagement trajectory into linear measurements of $\theta_a$, so the posterior over latent quality updates in closed form as censored outcomes arrive. Thompson sampling draws from that posterior to choose items, and the regret analysis uses inverse-propensity-weighted posterior covariances to show that the variance at the optimal arm is controlled by the full-feedback covariance multiplied by $\exp(-2\,\mathrm{VoPF}(t,Z_a))$.

What would settle it

Run the fitted-prior algorithm on a held-out domain where early engagement is uncorrelated with the 60-day reward (estimated VoPF near zero), and compare the log-ratio of delayed versus progressive regret to the predicted $\exp(-\mathrm{VoPF}(t))$ curve; if regret does not improve before day 60, or if the measured ratio trails the prediction, the central pre-reward learning claim is contradicted.

Watch

Extended reading notes

Core claim

The paper's central claim is that with delayed long-term rewards and progressive feedback, a Thompson-sampling algorithm driven by a Bayesian filter can achieve low expected regret before any long-term reward is observed, at a rate governed by the Value of Progressive Feedback. Formally, under a Gaussian model in which each item has a latent quality vector $\theta_a$ and user trajectories are i.i.d. noise around it, the expected instantaneous regret at batch $t$ of the lower-variance Impatient Bandit is bounded by $\exp(-\mathrm{VoPF}(t))\cdot \sigma_R\sqrt{2|A|\log(|A|)/(\sigma_R^2(r_1^\top \Sigma_1 r_1)^{-1}+|U_{<(t-d_{\max})}|)} + O(\epsilon_{\mathrm{rnd}})$; the square-root factor is the regret of Thompson sampling that waits for delayed rewards, and $\mathrm{VoPF}(t)$ is the conditional mutual information between the mean reward and the progressive observations given only the fully delayed observations. The paper further claims that in a large podcast-recommendation A/B test, the progressive-feedback treatment raised discovery rate by roughly 30% and 60-day active days, minutes, and return days per impression by over 50% for recently released shows relative to the delayed-reward control, and that the method was subsequently used in production.

Load-bearing premise

The posterior updates and the regret bound are only valid if the Gaussian prior learned from historical data correctly describes newly released items; the paper states in Section 9 that it provides no guarantee for when such a prior can be accurately learned.

Editorial extensions

If this is right

  • Before any long-term reward is fully observed, the algorithm can already achieve low expected regret in every batch when early engagement is informative, because the regret bound shrinks as $\exp(-\mathrm{VoPF}(t))$.
  • With uninformative progressive feedback the bound reduces to standard delayed-reward Thompson sampling; with perfect leading indicators it matches no-delay Thompson sampling, so the result interpolates between those extremes.
  • In the podcast simulations, the progressive-feedback policy accumulated far less regret than the delayed-reward policy during the first 60 days, and in a continuously changing action set the delayed policy never improved while the progressive policy kept cumulative regret under half.
  • In the reported A/B test, recently released shows recommended by the treatment policy had roughly 30% higher discovery rate and over 50% higher 60-day active days, minutes, and return days per impression compared with the delayed-reward control.
  • Since $\mathrm{VoPF}$ is a property of the data-generating environment, it can be estimated from historical data before deployment and used to compare candidate progressive feedback signals.

Reading between the lines

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

  • Editorial inference: a natural logged-data test of the theory is to estimate $\mathrm{VoPF}(t)$ from the fitted prior and compare it with the measured log-ratio of delayed versus progressive regret; a mismatch would diagnose prior misspecification or non-stationarity.
  • Editorial inference: the mechanism transfers to other delayed-outcome decisions with intermediate signals—clinical trials with biomarker surrogates, marketing with early click signals, or content moderation with early user reports—wherever a Gaussian filter can be fit from historical trajectories.
  • Editorial inference: the acknowledged reliance on a correctly learned prior means the pre-reward gains may shrink for genuinely novel content categories with no historical analogues, and the A/B-test gains should not be assumed to persist when the item distribution shifts.
  • Editorial inference: because VoPF is algorithm-independent and computable from data, it could be used prospectively to screen which surrogate signals justify an expensive online experiment before committing to deployment.
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

4 major / 5 minor

Summary. The paper formalizes a multi-armed bandit setting with progressive feedback and delayed rewards, proposes an Impatient Thompson Sampling algorithm built on a Gaussian Bayesian filter, and proves a per-batch regret bound whose improvement over delayed-reward Thompson sampling is governed by an information-theoretic quantity called the Value of Progressive Feedback (VoPF). The authors fit the Gaussian prior by empirical Bayes, validate the method on synthetic and Spotify podcast data, and report a large-scale Spotify A/B test in which a deterministic ranking variant with progressive stickiness estimates improves long-term engagement metrics, especially for recently released shows.

Significance. If the main results hold, this is a valuable contribution: VoPF is an environment-dependent, algorithm-independent quantity, and the regret bound interpolates cleanly between delayed and immediate feedback. The proof machinery, which builds on inverse-propensity weighting and lemmas from Qin and Russo, is substantive and mostly carefully presented. The synthetic experiments qualitatively match the theoretical prediction, and the large-scale A/B test is rare and valuable industrial evidence for the practical value of progressive feedback. The main limitations are that the theoretical guarantees require a correctly specified Gaussian prior and that the A/B test evaluates a deterministic ranking variant rather than the proposed bandit algorithm itself; the manuscript's broadest claims are stronger than what the evidence supports in these two respects.

major comments (4)
  1. [Theorem 1; §7.4; §9] The central regret guarantee is derived under Assumptions 1–4, which include a Gaussian outcome model and a prior distribution for theta_a that is known to the algorithm. Section 9 explicitly disclaims any guarantee about learning an accurate prior from previous data. Section 7.4 (Figure 7) then shows that replacing the fitted prior with an isotropic uninformative prior makes Progressive essentially match Delayed, so the exp(-2·VoPF) improvement is not a property of progressive feedback alone; it is jointly mediated by the quality of the empirical-Bayes prior. Because Proposition 2 and Lemmas 2–5 are exact only under this correctly specified Gaussian model, the paper's practical claim that the algorithm 'significantly outperforms' delayed-reward methods is not established under prior misspecification. Please add a misspecification or robustness analysis, or an empirical-Bayes consistency result with explicit rates, and qualify the abstract and Section 1 claims accordingly.
  2. [§8.4 and Section 8 introduction] The A/B test does not evaluate Algorithm 1 or Algorithm 2. Section 8 states that all tested variants use deterministic ranking strategies, and Section 8.4 specifies that the treatment policy replaces the delayed stickiness estimate with the posterior mean from the Gaussian filter, with no Thompson-sampling exploration. The reported gains therefore validate the progressive-feedback filtering model in a ranking system, but not the bandit exploration or the regret guarantees that constitute the paper's theoretical contribution. The abstract's statement that the approach is 'demonstrated by an A/B test' should either be scoped to this deterministic ranking variant or be accompanied by an online experiment that actually runs the bandit algorithm.
  3. [§8.5, Figure 11, Table 1] The headline empirical claims—over 50% increases on three recent-show metrics and 7–10% increases on all-show metrics—are supported only by point estimates with standard errors. The manuscript does not report the sample sizes behind recent-show impressions, confidence intervals for the reported ratios, or any multiple-comparison control across the four primary metrics. Since these numbers are central to the paper's applied significance, please provide formal interval estimates or a pre-specified analysis plan; if the standard errors shown are meant to be sufficient, the authors should explain why they support the stated conclusions.
  4. [Assumption 3; §6.1] The formal model assumes Gaussian engagement outcomes (Assumption 3), but the Spotify data used in Sections 6.1, 7.3–7.5, and 8 consist of binary daily indicators Y_u^(j) = 1{user engaged on day j}. This is a model misspecification in the main empirical application, so the exact Gaussian posterior and the regret bound do not directly cover the data-generating process used in the real-data experiments. The paper should state this mismatch as a limitation and provide calibration or robustness evidence, such as posterior predictive checks or a comparison to a non-Gaussian filter, rather than presenting the Gaussian filter as the model for the real data.
minor comments (5)
  1. [§4, Algorithm 1 description] In the paragraph following Algorithm 1, the phrase 'the posterior update rule used in line 11 of Algorithm 1 is can be derived' contains a grammatical typo; please revise.
  2. [Appendix D.2, Lemma 5 proof] In the proof of Lemma 5, the second entropy term uses \tilde Y_{u,t-1}(a) in one display, while the statement and the rest of the proof use \tilde Y_{u,t}(a); this should be corrected to be internally consistent.
  3. [§7.2] The text says 'reduce the number of actions per round dramatically from m = 1000 to m = 10,' but m is the batch size, not the number of actions per round; please rephrase to avoid confusion.
  4. [Figure 3] The y-axis label 'Information entropy [Nats]' is misleading because the plotted quantity is the Value of Progressive Feedback measured in nats, not an entropy; please relabel the axis accordingly.
  5. [§8.5] The phrase 'huge increases' is informal for a refereed paper; please replace it with precise effect sizes and confidence intervals, and reserve magnitude language for the discussion section.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: VoPF is an environment-defined information quantity and the regret bound is derived from posterior variance identities rather than fitted to the data.

full rationale

The central regret bound (Theorem 1, Corollary 1) is not circular. VoPF is defined as a conditional mutual information (Eqs. 7 and 11) that is a property of the data-generating environment, not a fitted constant; the paper derives the exp(-VoPF(t)) improvement factor through Lemma 5, which rewrites VoPF as a log ratio of posterior variances, followed by a posterior-covariance comparison (Lemma 3) that is an external, parameter-free mathematical lemma from Qin and Russo [2023]. The 'delayed rewards' baseline term is the posterior standard deviation under delayed observations, computed from the same Gaussian model, so the bound is an algebraic consequence of the stated assumptions rather than an assumption of the conclusion. The empirical-Bayes prior is an input to the model, and the paper explicitly acknowledges in Section 9 that it provides no guarantees about learning an accurate prior; Section 7.4 further shows that an uninformative prior removes the progressive-feedback advantage. This is an honest scope condition and a misspecification risk, not a circular step. Self-citations to McDonald et al. [2023] for the algorithm origin and to Maystre et al. [2023] for the control policy are disclosed, and the present theoretical claims are proved in the text or via independent lemmas; the A/B test is an external empirical comparison. No derivation step reduces to its own input.

Assumptions & free parameters 3 free parameters · 6 assumptions · 2 invented entities

The central claim rests on exchangeable users, i.i.d. items, Gaussian outcomes, a correctly learned empirical-Bayes prior, and a rounding scheme. The Gaussian assumption is violated by the binary engagement data used in the Spotify experiments, and the prior-learning assumption is explicitly unproved. VoPF is a defined quantity, not a fitted parameter.

free parameters (3)
  • Prior mean vector µ1,z for each feature class z = Estimated from historical podcast data via type-II maximum likelihood (Section 6)
    The posterior beliefs and Thompson sampling decisions depend on this prior mean; the paper provides no guarantee that the fitted prior is accurate for new items.
  • Prior covariance matrix Σ1,z for each feature class z = Estimated as the empirical covariance of show-level mean engagement vectors (Section 6)
    Controls the information content of the prior and the VoPF; misspecification would bias the Bayesian filter and invalidate the regret bound.
  • Noise covariance matrix Vz = Estimated by averaging per-show residual covariance matrices (Section 6)
    Determines how much weight progressive observations receive in the posterior update; in practice the data are binary while the model assumes Gaussian noise.
assumptions (6)
  • domain assumption Exchangeability of potential outcomes (Assumption 1)
    Users are exchangeable conditional on item features, enabling a De Finetti representation and i.i.d. outcomes given θa. It is a behavioral modeling assumption that underpins the posterior filter.
  • domain assumption Randomly drawn items (Assumption 2)
    Items and their latent features are i.i.d., which defines the Bayesian regret and the prior over θa. Needed for the expected regret to average over item draws.
  • domain assumption Gaussian potential outcomes and affine reward (Assumption 3)
    Gaussian process outcomes and a linear reward function yield closed-form Kalman-style posterior updates and are used throughout Lemma 1, Proposition 2, and Theorem 1. The Spotify application uses binary daily indicators, so this assumption is violated in practice.
  • ad hoc to paper Rounding procedure (Assumption 4)
    Bounds the error from rounding Thompson sampling probabilities to integer user allocations; Algorithm 6 is shown to satisfy it. Needed for the O(ϵ_rnd) terms in the regret bounds.
  • standard math De Finetti's theorem
    Used in Lemma 1 to justify the existence of the latent item variable θa and the conditional i.i.d. structure of user outcomes.
  • standard math Gaussian entropy and scalar posterior variance formulas
    Used in Lemma 5 to express VoPF as a log-ratio of variances and in the proof of Theorem 1 to simplify the delayed-feedback variance terms.
invented entities (2)
  • Value of Progressive Feedback (VoPF) independent evidence
    purpose: Information-theoretic quantity in the regret bound; measures how much early engagement outcomes reduce entropy of the expected long-term reward relative to delayed outcomes alone.
    Defined via mutual information in Eq. (7). It is a falsifiable construct: Section 7.2 compares the theoretically computed VoPF to the empirical log regret ratio and finds qualitative agreement.
  • Item latent quality θa
    purpose: Represents persistent unobserved item quality; conditional on θa and features, user outcomes are i.i.d.
    A standard latent variable introduced through the De Finetti representation of exchangeable outcomes. It is not directly measured, but its existence follows from Assumption 1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Impatient Bandits: Optimizing for the Long-Term Without Delay." pith.science (2026). https://pith.science/paper/STAOABRL

@misc{pith2026250107761,
  author       = {Pith},
  title        = {Pith review of: Impatient Bandits: Optimizing for the Long-Term Without Delay},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/STAOABRL}},
  note         = {Machine review of arXiv:2501.07761}
}
read the original abstract

Increasingly, recommender systems are tasked with improving users' long-term satisfaction. In this context, we study a content exploration task, which we formalize as a bandit problem with delayed rewards. There is an apparent trade-off in choosing the learning signal: waiting for the full reward to become available might take several weeks, slowing the rate of learning, whereas using short-term proxy rewards reflects the actual long-term goal only imperfectly. First, we develop a predictive model of delayed rewards that incorporates all information obtained to date. Rewards as well as shorter-term surrogate outcomes are combined through a Bayesian filter to obtain a probabilistic belief. Second, we devise a bandit algorithm that quickly learns to identify content aligned with long-term success using this new predictive model. We prove a regret bound for our algorithm that depends on the Value of Progressive Feedback, an information-theoretic metric that captures the quality of short-term leading indicators that are observed prior to the long-term reward. We apply our approach to a podcast recommendation problem, where we seek to recommend shows that users engage with repeatedly over two months. We empirically validate that our approach significantly outperforms methods that optimize for short-term proxies or rely solely on delayed rewards, as demonstrated by an A/B test in a recommendation system that serves hundreds of millions of users.

Figures

Figures reproduced from arXiv: 2501.07761 by the authors.

Figure 1
Figure 1. Short-term outcomes enable a rapid feedback loop, but might be poorly aligned with long-term success metrics, which take longer to realize. Our method bypasses this apparent tradeoff by adaptively making use of all available information at a given time. On these large-scale platforms, algorithms need to make decisions at a rapid pace, since the time between the start of one user’s interaction and the start of the ne… view at source ↗
Figure 2
Figure 2. Recent efforts to optimize for longer-term metrics, such as user engagement over a 60-day [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 2
Figure 2. Progressive engagement feedback from two hypothetical users. The long-term outcome of [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (10 more)
Figure 3
Figure 3. Figure 3: Prior learned on Spotify interaction data. Left: predictive error of the reward model as [PITH_FULL_IMAGE:figures/full_fig_p019_3.png]
Figure 4
Figure 4. Figure 4: Cumulative regret of for three different settings of the generative model, averaged over [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: A comparison between the Value of Progressive Feedback, a theoretical quantity ( [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Cumulative regret of different bandit algorithms on the Spotify problem with [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Cumulative regret of two variants of our bandit algorithm using different prior and [PITH_FULL_IMAGE:figures/full_fig_p023_7.png]
Figure 8
Figure 8. Figure 8: Cumulative regret of different bandit algorithms on a variant of the Spotify problem, [PITH_FULL_IMAGE:figures/full_fig_p024_8.png]
Figure 9
Figure 9. Figure 9: Cumulative regret of our Thompson Sampling Impatient Bandit algorithm (solid lines) [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]
Figure 10
Figure 10. Figure 10: Screenshots of Spotify mobile application. The highlighted podcast discovery shelf contains up to 20 cards that the user can scroll through. The methodology we describe now governs podcast rec￾ommendations in several parts of the app, but we focus here on results from…
Figure 11
Figure 11. Figure 11: Results of treatment policy compared to control policy for recommending all shows [PITH_FULL_IMAGE:figures/full_fig_p029_11.png]
Figure 12
Figure 12. Figure 12: Stickiness of discoveries of recently released shows. [PITH_FULL_IMAGE:figures/full_fig_p030_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Bandit Simulation for Average Reward Inference

    stat.ML 2026-05 unverdicted novelty 7.0 of 10

    BSI fits an environment simulator from bandit data and propagates parameter uncertainty to produce asymptotically valid confidence intervals for mean reward under arbitrary evaluation policies, including black-box ada...

Reference graph

Works this paper leans on

13 extracted references · 11 canonical work pages · cited by 1 Pith paper

  1. [1]

    context feature vectors

    (18) Thus we have shown that, with probability 1,θa ≜ lim|U |→∞ ¯YU (a) exists and is equal toE Yu(a) | Pa . Almost sure convergence implies convergence in distribution. Therefore,lim|U |→∞ L( ¯YU (a) | Za = z) = L(θa | Za = z) where L(X) denotes the law (or probability distribution) of a random variable X. To show(16), we calculateL( ¯YU (a) | Za = z) an...

    work page 2009

  2. [2]

    de-correlated

    Using the matrixLz (recall Lz is lower triangular andVz = LzL⊤ z ), we can define a version ofYu, whose noise covariance matrix is “de-correlated”: Yu ≜ L−1 z Yu = L−1 z  Yu −   X ⊤ u θ(1) a ... X ⊤ u θ(J) a     + L−1 z   X ⊤ u θ(1) a ... X ⊤ u θ(J) a   . Above θa | (Xu, Za = z) ∼ N (µ1,z, Σ0,z) and L−1 z  Yu −   X ⊤ u θ(1) a ... X...

    work page 1970

  3. [3]

    Kallus and X

    N. Kallus and X. Mao. On the role of surrogates in the efficient estimation of treatment effects with limited outcome data.arXiv preprint arXiv:2003.12408,

    arXiv 2003

  4. [4]

    full feedback

    D.2 Helpful Lemmas By Lemma 2, in order to bound regret, it is enough to boundEπTS-rnd P a∈A p∗ t,aσ2 t,a . Note, EπTS-rnd X a∈A p∗ t,aσ2 t,a = EπTS-rnd X a∈A 1 A∗=aσ2 t,a = EπTS-rnd X a∈A 1 A∗=ar⊤ 1 Cov (θa | Ht) r1 . Furthermore, note that Cov (θa | Ht) = Σ−1 1,Za + X u∈U<t 1 Au=a JX j=1 1 {t > τu + dj}ℓ(j) Za (ℓ(j) Za )⊤ −1 = Σ−1 1,Za + m t−1X t′=1 pTS...

    work page 2023

  5. [6]

    Q. Wu, H. Wang, L. Hong, and Y. Shi. Returning is believing: Optimizing long-term user engagement in recommender systems. InProceedings of the 2017 ACM on Conference on Information and Knowledge Management, CIKM ’17,

    work page 2017

  6. [12]

    Below, we consider anyu ∈ Ut. By the result above, we have that E ∆t(πTS-rnd) | Ht = X a∈A pTS-rnd t,a E R Yu(A∗) − R Yu(a) | Ht = X a∈A pTS t,a E R Yu(A∗) − R Yu(a) | Ht + X a∈A pTS-rnd t,a − pTS t,a E R Yu(A∗) − R Yu(a) | Ht ≤ X a∈A pTS t,a E R Yu(A∗) − R Yu(a) | Ht + X a∈A pTS-rnd t,a − pTS t,a E R Yu(A∗) − R Yu(a) | Ht Above recall thatpTS-rnd t,a is ...

    work page 2023

  7. [2013]

    Anderer, H

    A. Anderer, H. Bastani, and J. Silberholz. Adaptive clinical trial designs with surrogates: When should we bother?Management Science, 68(3):1982–2002,

    work page 1982

  8. [2015]

    Optimizing Audio Recommendations for the Long-Term: A Reinforcement Learning Perspective

    L. Maystre, D. Russo, and Y. Zhao. Optimizing audio recommendations for the long-term: A reinforcement learning perspective.arXiv preprint arXiv:2302.03561,

    work page Pith review arXiv

Show all 13 references
  1. [2019]

    Let Assumptions 1 and 3 hold

    35 A Multivariate Gaussian Bayesian Model: Proof of Lemma 1 Lemma 1 (Multivariate Gaussian Bayesian Model). Let Assumptions 1 and 3 hold. The limit θa ≜ lim|U |→∞ 1 |U | P u∈U Yu(a) exists almost surely. Conditional on θa, Za, the random variables {Yu(a)}u∈U are i.i.d. Moreove...

    1976

  2. [2020]

    Baran, G

    B. Baran, G. D. Junior, A. Danylenko, O. S. Folorunso, G. Forsum, M. Lefarov, L. Maystre, and Y. Zhao. Accelerating creator audience building through centralized exploration. InProceedings of RecSys 2023, Singapore, Sept

    2023

  3. [2021]

    Zheng, F

    G. Zheng, F. Zhang, Z. Zheng, Y. Xiang, N. J. Yuan, X. Xie, and Z. Li. Drn: A deep reinforcement learning framework for news recommendation. In Proceedings of the 2018 World Wide Web Conference, WWW ’18,

    2018

  4. [2022]

    doi: 10.1214/22-AOS2191. Z. Liu, S. Liu, Z. Zhang, Q. Cai, X. Zhao, K. Zhao, L. Hu, P. Jiang, and K. Gai. Sequential recommendation for optimizing both immediate feedback and long-term retention. InProceedings of the 47th International ACM SIGIR Conference on Research and Deve...

    doi

  5. [2024]

    Zhang, M

    V. Zhang, M. Zhao, A. Le, and N. Kallus. Evaluating the surrogate index as a decision-making tool using 200 a/b tests at netflix.arXiv preprint arXiv:2311.11922,

    arXiv

Pith tools

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