Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Separating immediate engagement from 14-day-delayed sparse actions improves live-streaming ranking where unified models degrade: +0.09% DAV, +0.27% follows, +0.56% dedicated ARPU.

desk verdict A solid industrial case study from Twitch: the fresh/delayed signal separation finding is believable, but the reported effect sizes are tuned on the same metrics they are measured on, so treat the numbers as conditional. read the letter →

arxiv 2608.04455 v1 pith:4JUYPDZP submitted 2026-08-05 cs.IR cs.LG

Xiaoyi Gu , Julia Tavares , Eder Santana , Carlos Mendoza-Cardenas , Nikita Mishra , Saad Ali This is my paper
classification cs.IRcs.LG
keywords multi-objectiverankinglive-streamingrecommendationdelayedfeedbacktargetsparsityviewersegmentbiasmixture-of-expertsfreshandsignalssystem
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 argues that for live-streaming recommendation, the right response to sparse and delayed viewer behavior is to separate signals by time scale rather than to model all objectives in one network. It proposes a delayed-window construction that labels sparse actions (chat, follow, spend) as positive if they occur within 14 days of an impression, while keeping immediate short-watch engagement in a fresh-signal model, and it combines the predictions through segment-specific weights at serving time. The final system, FSM-MMoE-VST, jointly models deep engagement and delayed actions in a multi-gate mixture-of-experts network. In staged 14-day A/B tests, the MMoE enhancement added +0.08% overall DAV, +0.27% follows, and +0.10% long-view engagement, atop a baseline that had already won a previous revenue optimization; the delayed-window multi-model alone added +0.09% DAV and +0.56% capped ARPU among dedicated viewers. A reader should care because this is a practical recipe for balancing engagement, retention, and monetization when positive actions are rare and arrive late.

What carries the argument

The load-bearing object is the delayed-window indicator $y^{\mathrm{delayed}}_{v_i,c_j} = \mathbb{I}[|DW_{v_i,c_j}(t, \Delta t)| > 0]$, where $DW$ collects actions on channel $c_j$ by viewer $v_i$ in the 14 days after exposure; this converts rare chat, follow, and spend events into denser binary targets for the delayed-signal models. Around this sit two further mechanisms: the fresh/delayed model split, which keeps dense immediate short-watch signals out of the same network as sparse delayed targets, and the segment-aware weighting in Eq. (5), which adjusts action weights for Early versus Dedicated viewers at serving time without retraining. MMoE supplies the joint tower for deep engagement and delayed actions; the ablations show that the split, not the multi-task backbone, carries the gains.

What would settle it

Re-build the sparse-target labels so that each delayed action is attributed only to exposures on the same recommendation surface (or to the last exposure before the action) and rerun the offline NDCG evaluation and the 14-day online A/B; if the DAV, follow, and capped-ARPU gains disappear or reverse, the 14-day attribution window, not the fresh/delayed model split, is what carries the improvement.

Watch

Extended reading notes

Core claim

The central claim is that the architectural separation of fresh and delayed signals is the primary driver of improvement in live-streaming recommendation, not the choice of multi-task backbone. The paper shows that any single unified model that jointly learns short watch (SMP) together with long watch, chat, follow, and spend degrades LMP NDCG@6 by roughly 4.2% to 4.6% regardless of backbone, whereas pairing an independent fresh-signal model for SMP with a delayed-signal model for sparse actions preserves engagement and improves spend. The final ranking function, $F_{\mathrm{FSM-MMoE-VST}}(x_{v_i,c_j}, s) = w_{\mathrm{smp},s}\, p^{\mathrm{FSM}}_{\mathrm{smp}}(x_{v_i,c_j}) + \sum_{a \in \{\mathrm{LMP}, \mathrm{chat}, \mathrm{follow}, \mathrm{spend}\}} w_{a,s}\, p^{\mathrm{MMoE}}_a(x_{v_i,c_j})$, applies segment-conditioned weights at inference so early viewers are pushed toward immediate watch behavior and dedicated viewers toward deeper engagement and monetization. Online, the stagewise additions produced +0.09% DAV and +0.56% dedicated-viewer capped ARPU for the multi-model with delayed window, +0.15% early-viewer DAV for segment targeting, and +0.08% overall DAV and +0.27% follows for the MMoE enhancement, while MMoE cut delayed-target parameters by 41.9%.

Load-bearing premise

The delayed-window labels in Eq. (2)-(3) attribute every chat, follow, or spend action that happens within 14 days of an impression to that impression—even when the viewer may have reached the channel through another surface, as the paper itself notes in Section 5.2.1—so if that attribution is biased toward popular channels or recent exposures, the sparse-target models learn from mislabeled positives and the measured online gains are overstated.

Editorial extensions

If this is right

  • If the separation of fresh and delayed signals is the dominant factor, then teams working with sparse delayed targets should first split models by signal density and time scale before choosing a multi-task backbone.
  • The 14-day window is a tunable parameter with diminishing returns: the paper finds longer windows stop improving spend and start degrading engagement, so other services can calibrate the same trade-off with their own NDCG curves.
  • The same multi-model architecture transferred to a second surface, the mobile live feed, where a 14-day A/B test saw +1.12% positive user-channel interactions.
  • Consolidating the delayed-target models into a single MMoE cuts parameters by 41.9% while improving follows and DAV, keeping the approach within the sub-110ms p99 latency budget for real-time serving.
  • Because the +0.56% capped ARPU gain among dedicated viewers was measured against a baseline that already included a monetization heuristic, the delayed-window modeling rather than the removal of a weak heuristic carries the revenue improvement.

Reading between the lines

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

  • The paper leaves an attribution question open: the 14-day window credits any qualifying action to the preceding impression even when other surfaces or earlier exposures contributed, so a same-surface attribution test would show how much of the +0.09% DAV gain is genuine signal.
  • If backbone choice really is secondary once fresh and delayed signals are separated, the same design should transfer to other live-streaming or short-video services that prefer cheaper MTL architectures than MMoE.
  • The E/D segment weights are hand-tuned; an online-learning variant that adapts them from user response is a natural extension, and if learned weights outperform the fixed ones, the reported gains are a lower bound.
  • The comparison with a monetization heuristic suggests a broader design lesson: replacing a revenue feature bolted onto an engagement model with an explicitly modeled delayed spend target can improve dedicated-segment revenue without sacrificing engagement.
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 describes a multi-objective ranking system for Twitch live-streaming recommendations, addressing sparse and delayed user actions through three main components: a delayed-window label construction for sparse targets (chat, follow, spend), a multi-model architecture that separates fresh signals (SMP) from delayed signals, and a segment-aware inference-time weighting module (VST) that later is combined with an MMoE backbone for the delayed/deeper targets. The authors report offline NDCG@6 comparisons across architectural variants, three staged 14-day online A/B experiments, an ablation of MMoE task grouping, and a generalization test on the Twitch mobile feed. The headline results are small but statistically significant online gains, e.g., +0.09% DAV for the multi-model, +0.15% E-viewer DAV for VST, and +0.08% DAV / +0.27% follows for the MMoE enhancement, alongside a 41.9% parameter reduction and sub-110ms p99 latency.

Significance. If the results hold, the paper provides a useful industrial case study showing that separating fresh and delayed signals can matter more than the choice of multi-task backbone in live-streaming recommendation. The staged online A/B testing on live traffic, the use of CUPED, the parameter-reduction claim, and the independent validation on the Twitch mobile feed are genuine strengths that lend credibility to the engineering contribution. The main limitations are that offline gains are extremely small and unreplicated, several key choices (delayed window length, VST weights) were selected using the same metrics later presented as evidence, and the delayed-window label definition may misattribute actions to impressions without controlling for other exposures.

major comments (4)
  1. [§4.2, Eqs. (2)–(3), §5.2.1] The delayed-window labels defined in Eqs. (2)–(3) assign a positive to every impression for which any action on channel c_j occurs within Δt days, with no correction for other exposures to the same channel through search, browse, or other recommendation surfaces. The paper itself acknowledges in §5.2.1 that "excessive window lengths introduce noise by attributing actions to impressions they are no longer associated with," but it never quantifies this attribution error at the chosen 14-day window. For popular channels, a large fraction of within-window actions may occur regardless of the specific impression, so the DSM/MMoE targets may contain many misattributed positives; the reported offline and online gains would then reflect popularity rather than genuine delayed feedback. To make the paper's central claim load-bearing, please add a robustness analysis, for example comparing models trained on single-exposure versus multi-exposure impressions, controlling for exposure count, or evaluating label precision on a manually or last-touch-annotated subset.
  2. [§5.2.2, Table 1] The offline gains are very small and are reported without any measure of uncertainty. For instance, the final FSM+MMoE+VST LMP NDCG@6 improves from 0.2459 to 0.2462 and differs from FSM+Shared-Bottom+VST by 0.0001; differences of this size can easily arise from random validation noise, and the choice of the 14-day window and VST weights was made on these same metrics. Please report standard errors or confidence intervals, ideally across multiple validation days or bootstrap resamples, and state how many seeds or checkpoints were used. The conclusion in §7.1 that signal separation rather than MTL backbone drives gains is supported by near-null differences that need an explicit power or equivalence argument.
  3. [§5.3, Exp. 1] Experiment 1 simultaneously introduces the 14-day delayed window and the multi-model separation relative to the baseline, so the measured gains cannot isolate the mechanism claimed in §7.1. The baseline is also a single-objective SMP model with a MAR heuristic, meaning the treatment replaces a heuristic monetization term with learned delayed-target models. Please provide a factorized comparison (e.g., single model with delayed labels versus MM with delayed labels, or MM with immediate sparse labels versus MM with delayed labels) either offline or in a smaller online test, so that the "separation is the primary driver" claim is not confounded by the label change and the baseline heuristic.
  4. [§5.1.3, §5.2.1, §5.3 Exp. 2] The VST weights are "selected candidate weights in offline tests and finalized ... through iterative A/B testing," and the 14-day window is selected from the same offline NDCG curves that later appear as evidence. Consequently, the reported improvements, especially Exp. 2's +0.15% E DAV and Exp. 3's +0.08% DAV, are partly an in-sample selection result. The staged A/B design is a strength, but the paper should clarify which choices were fixed before the reported experiments and provide a validation on a holdout period or pre-registered weights to separate selection from true improvement.
minor comments (5)
  1. [§5.1.2] The evaluation dataset uses a 35-day forward window for ground-truth labels while training uses 14-day delayed windows; this discrepancy is not explained and affects how the offline NDCG numbers should be interpreted. Please clarify whether the 35-day labels were used for all targets in Table 1 and Figure 3, and discuss the implications.
  2. [Abstract, §1, §8] The phrase "millions more annual active viewer days" is an extrapolation from a +0.09% DAV lift, but no baseline DAV level or calculation is provided. Please state the derivation or remove the extrapolation.
  3. [Table 2] Each experiment tests multiple metrics across multiple segments, and Exp. 3 reports p-values for several endpoints; the paper does not discuss multiple-comparison control or report confidence intervals. Adding CIs or a note on the number of pre-specified metrics would help readers calibrate the significance claims.
  4. [Figure 3] The E-viewer Spend NDCG@6 axis spans only 0.0308–0.0311, making differences across window lengths appear larger than they are; please use a consistent scale across panels or state that the axis range is intentionally narrow.
  5. [Table 1] The table would be easier to use if the standard deviations across random initializations or validation days were included, and if the "All" column made explicit that D viewers dominate the aggregate given the stated 4x sample imbalance.

Circularity Check

1 steps flagged · score 4.0 of 10

VST weight tuning on the reported DAV/LMP metrics makes the +0.15% E-DAV gain partly a selection artifact; the core fresh/delayed signal separation claim remains externally validated.

  1. fitted input called prediction [Section 5.1.3 and Eq. (4); reported in Section 5.3, Experiment 2, Table 2]
    ""We selected candidate weights in offline tests and finalized the production configuration through iterative A/B testing on business metrics and guardrails." ... "Our segment-aware approach successfully improves E viewer engagement (+0.15% DAV, +0.25% LMP) while maintaining D performance in the A/B (Table 2, Exp. 2) with the best-performing weights.""

    The VST ranking score in Eq. (4) is a weighted sum whose tunable parameters are the segment-conditioned weights w_{a,s}. The paper states these weights were finalized through iterative A/B testing on business metrics, and the headline result of Experiment 2 is the improvement on those same business metrics (DAV and LMP) for E viewers, obtained 'with the best-performing weights.' Selecting the weight configuration that maximizes the reported DAV/LMP outcomes and then presenting the selected configuration's DAV/LMP lift as the result makes the reported gain the maximum of the selection distribution rather than an out-of-sample prediction.

full rationale

Aside from the VST weight-selection issue, the paper's central derivation chain is largely self-contained and externally validated. The delayed-window label construction in Eqs. (2)-(3) is a modeling definition, not a quantity derived from the claimed outcomes; the 14-day window length is chosen offline via NDCG curves and then tested online on fresh traffic. Experiment 1 tests the multi-model with delayed targets against a prior A/B-winning baseline, Experiment 2 adds VST, and Experiment 3 replaces independent DSMs with MMoE, giving a staged decomposition. The mobile-feed A/B with +1.12% positive interactions provides an independent validation surface. The only author-overlapping citation, [6], describes the pre-existing retrieval stage and is not load-bearing for the paper's main claims. The acknowledged attribution bias for long delayed windows ('attributing actions to impressions they are no longer associated with') is a correctness risk, not a circular step, because the online metrics are real product outcomes rather than the delayed-window labels themselves. Overall, the core architectural claim does not reduce to its inputs by construction; only the VST improvement is partly a selection artifact, warranting a moderate score.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a set of tunable weights, thresholds, and window lengths that are not derived from first principles. The most consequential are the segment-conditioned weights and the 14-day window, both selected using the same business metrics the paper reports as evidence. The action-to-objective mapping and the E/D segmentation are domain assumptions inherited from product definitions.

free parameters (7)
  • Delayed window length Δt = 14 days
    Selected by comparing NDCG@6 across {7,14,21,28,35} days (Section 5.2.1, Figure 3), then used in online A/B. 21 days was similar but rejected for computation/staleness.
  • Segment-conditioned weights w_{a,s} = Proprietary
    Finalized via offline tests and iterative online A/B (Section 5.1.3); these weights directly determine the reported VST gains, so they are fit to the outcome metric.
  • Global objective weights w_a = Proprietary
    Appear in Eq. 1 and are part of the ranking scalarization; tuning details not disclosed.
  • Segment thresholds M and N = Proprietary
    Define Early vs Dedicated viewers (Section 3.2); the whole segment-targeting claim depends on this split.
  • Watch-time thresholds τ_s and τ_l = Proprietary
    Define short vs long minutes play (footnote 5); the target definitions rely on them.
  • Number of MMoE experts K = 4
    Selected via hyperparameter optimization (Section 4.5); affects capacity of the joint model.
  • Learning rates and batch sizes = LR in [0.001, 0.005, 0.01], batch in [1024, 2048]
    Selected through HPO (Section 5.1.2).
assumptions (5)
  • domain assumption The five viewer actions (SMP, LMP, chat, follow, spend) are valid proxies for engagement, retention, and monetization objectives
    Section 3.2 states this mapping between actions and business goals; the whole multi-objective setup relies on it.
  • domain assumption Early vs Dedicated segment split (defined by M and N thresholds) captures meaningful lifecycle differences that require different ranking objectives
    Section 3.2 and Section 4.4. No empirical validation of the threshold choice is provided beyond the downstream metrics.
  • domain assumption A 14-day attribution window is long enough to capture delayed feedback and short enough to avoid stale attributions
    Chosen based on offline NDCG; the paper acknowledges diminishing returns beyond 14 days (Sections 5.1.1, 5.2.1).
  • domain assumption The two-stage retrieval/ranking decomposition is fixed; ranking quality is measured only within retrieved candidates
    Section 3.1; improvements are conditional on the retrieval stage not being affected.
  • standard math Binary cross entropy and NDCG are appropriate learning and evaluation objectives for these imbalanced targets
    Section 5.1.2; this is a standard modeling choice, not derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting." pith.science (2026). https://pith.science/paper/4JUYPDZP

@misc{pith2026260804455,
  author       = {Pith},
  title        = {Pith review of: Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4JUYPDZP}},
  note         = {Machine review of arXiv:2608.04455}
}
read the original abstract

One of the most challenging problems entertainment live-streaming services face in recommendation systems is that user behaviors are sparse and delayed, and interaction data exhibits bias for different user segments. Unlike e-commerce applications where user actions follow linear sequences, live-streaming viewers engage in multiple concurrent behaviors of watching, chatting, following, and spending, each occurring with varying delays. We address these challenges through three key contributions: 1) a delayed window approach that extends feedback collection beyond immediate responses, 2) a multi-model architecture that combines fresh and delayed signals, and a segment-aware targeting module that optimizes ranking scores differently across user lifecycle stages, and 3) Multi-gate Mixture-of-Experts (MMoE) integration that jointly models correlated targets while reducing model parameters by 41.9% compared to independent models. Online A/B testing demonstrates significant improvements, including a +0.09% increase in Daily Active Viewers (DAV), generating millions more annual active viewer days, and +0.56% increase in highly engaged viewers' capped Average Revenue Per User (ARPU). Viewer-segment targeting achieved an additional +0.15% DAV improvement for newer and less engaged viewers, while MMoE enhancement added +0.08% overall DAV and +0.27% new follows. The proposed system processes ranking requests with low latency, providing a scalable approach for balancing multiple business objectives across diverse user populations. In addition, we tested the multi-model architecture on the Twitch mobile live feed and achieved a +1.12% increase in positive user-channel interactions (clicks, follows, and likes), demonstrating applicability beyond the primary use case.

Figures

Figures reproduced from arXiv: 2608.04455 by the authors.

Figure 1
Figure 1. Timeline illustration of Delayed Window approach. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The FSM-MMoE-VST architecture. FSM handles shallow engagement (SMP); MMoE jointly models deep engagement [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison of LMP and Spend NDCG@6 across delayed window sizes (1-35 days), using consistent multi-model [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 18 canonical work pages

  1. [1]

    Shaghayegh Agah, Shaun Schaeffer, Maria Peifer, Neeraj Sharma, Ankit Ma- heshwari, and Sardar Hamidian. 2025. Pareto-Optimal Solution: Optimizing Engagement and Revenue. InProceedings of the Nineteenth ACM Conference on Recommender Systems. 1034–1037

    work page 2025

  2. [2]

    Qingyao Ai, Keping Bi, Cheng Luo, Jiafeng Guo, and W Bruce Croft. 2018. Unbi- ased learning to rank with unbiased propensity estimation. InThe 41st interna- tional ACM SIGIR conference on research & development in information retrieval. 385–394

    2018

  3. [3]

    Yimeng Bai, Yang Zhang, Fuli Feng, Jing Lu, Xiaoxue Zang, Chenyi Lei, and Yang Song. 2024. GradCraft: Elevating Multi-task Recommendations through Holistic Gradient Crafting. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4774–4783

    work page 2024

  4. [4]

    Jiangxia Cao, Pengbo Xu, Yin Cheng, Kaiwei Guo, Jian Tang, Shijun Wang, Dewei Leng, Shuang Yang, Zhaojie Liu, Yanan Niu, et al. 2025. Pantheon: Personalized multi-objective ensemble sort via iterative pareto policy optimization. InPro- ceedings of the 34th ACM International Conference on Information and Knowledge Management. 5575–5582

    work page 2025

  5. [5]

    Olivier Chapelle. 2014. Modeling delayed feedback in display advertising. In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. 1097–1105

    2014

  6. [6]

    Edgar Chen, Mark Ally, Eder Santana, and Saad Ali. 2022. Weighing dynamic availability and consumption for Twitch recommendations. InKDD Workshop on Online and Adaptive Recommender Systems (OARS)

    work page 2022

  7. [7]

    Jiawei Chen, Hande Dong, Xiang Wang, Fuli Feng, Meng Wang, and Xiangnan He. 2023. Bias and debias in recommender system: A survey and future directions. ACM Transactions on Information Systems41, 3 (2023), 1–39

    2023

  8. [8]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. InProceedings of the 10th ACM conference on recommender systems. 191–198

    2016

Show all 42 references
  1. [9]

    Alex Deng, Ya Xu, Ron Kohavi, and Toby Walker. 2013. Improving the sensitivity of online controlled experiments by utilizing pre-experiment data. InProceedings of the sixth ACM international conference on Web search and data mining. 123–132

    2013

  2. [10]

    Ke Ding, Xin Dong, Yong He, Lei Cheng, Chilin Fu, Zhaoxin Huan, Hai Li, Tan Yan, Liang Zhang, Xiaolu Zhang, et al. 2021. MSSM: a multiple-level sparse sharing model for efficient multi-task learning. InProceedings of the 44th International ACM SIGIR Conference on Research and ...

    2021

  3. [11]

    Yuval Dishi, Ophir Friedler, Yonatan Karni, Natalia Silberstein, and Yulia Stolin

  4. [12]

    Zuohui Fu, Yikun Xian, Ruoyuan Gao, Jieyu Zhao, Qiaoying Huang, Yingqiang Ge, Shuyuan Xu, Shijie Geng, Chirag Shah, Yongfeng Zhang, et al. 2020. Fairness- aware explainable recommendation over knowledge graphs. InProceedings of the 43rd international ACM SIGIR conference on re...

    2020

  5. [13]

    Hui Gao and Yihan Yang. 2022. Multi-Head Online Learning for Delayed Feedback Modeling.arXiv preprint arXiv:2205.12406(2022)

    2022 arXiv

  6. [14]

    Olivier Jeunen, Jatin Mandav, Ivan Potapov, Nakul Agarwal, Sourabh Vaid, Wen- zhe Shi, and Aleksei Ustimenko. 2024. Multi-objective recommendation via multivariate policy learning. InProceedings of the 18th ACM Conference on Rec- ommender Systems. 712–721

    2024

  7. [15]

    Sofia Ira Ktena, Alykhan Tejani, Lucas Theis, Pranay Kumar Myana, Deepak Dilipkumar, Ferenc Huszár, Steven Yoo, and Wenzhe Shi. 2019. Addressing delayed feedback for continuous training with neural networks in CTR prediction. InProceedings of the 13th ACM conference on recomme...

    2019

  8. [16]

    Fengqi Liang, Baigong Zheng, Liqin Zhao, Guorui Zhou, Qian Wang, and Yanan Niu. 2024. Ensure timeliness and accuracy: A novel sliding window data stream paradigm for live streaming recommendation.arXiv preprint arXiv:2402.14399 (2024)

    2024 arXiv

  9. [17]

    Xiao Lin, Hongjie Chen, Changhua Pei, Fei Sun, Xuanji Xiao, Hanxiao Sun, Yongfeng Zhang, Wenwu Ou, and Peng Jiang. 2019. A pareto-efficient algo- rithm for multiple objective optimization in e-commerce recommendation. In Proceedings of the 13th ACM Conference on recommender sy...

    2019

  10. [18]

    Langming Liu, Wanyu Wang, Chi Zhang, Bo Li, Hongzhi Yin, Xuetao Wei, Wenbo Su, Bo Zheng, and Xiangyu Zhao. 2025. Multi-task Offline Reinforcement Learn- ing for Online Advertising in Recommender Systems. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery an...

    2025

  11. [19]

    Jiaqi Ma, Zhe Zhao, Jilin Chen, Ang Li, Lichan Hong, and Ed H Chi. 2019. Snr: Sub-network routing for flexible parameter sharing in multi-task learning. In Proceedings of the AAAI conference on artificial intelligence, Vol. 33. 216–223

    2019

  12. [20]

    Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. 2018. Modeling task relationships in multi-task learning with multi-gate mixture-of- experts. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1930–1939

    2018

  13. [21]

    Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. InThe 41st International ACM SIGIR Conference on Research & Development in Information Retriev...

    2018

  14. [22]

    Sofia Maria Nikolakaki, Siyong Ma, Srivas Chennu, and Humeyra Topcu Altintas

  15. [23]

    Sebastian Ruder. 2017. An overview of multi-task learning in deep neural net- works.arXiv preprint arXiv:1706.05098(2017)

    2017 arXiv

  16. [24]

    InProceedings of the Nineteenth ACM Conference on Recommender Systems

    SEMORec: A Scalarized Efficient Multi-Objective Recommendation Frame- work. InProceedings of the Nineteenth ACM Conference on Recommender Systems. 1074–1077

  17. [25]

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

    2016

  18. [26]

    Tobias Schnabel and Paul N Bennett. 2020. Debiasing item-to-item recommenda- tions with small annotated datasets. InProceedings of the 14th ACM Conference on Recommender Systems. 73–81

    2020

  19. [27]

    Trevor Standley, Amir Zamir, Dawn Chen, Leonidas Guibas, Jitendra Malik, and Silvio Savarese. 2020. Which tasks should be learned together in multi-task learning?. InInternational conference on machine learning. PMLR, 9120–9132

    2020

  20. [28]

    Chenglei Shen, Jiahao Zhao, Xiao Zhang, Weijie Yu, Ming He, and Jianping Fan

  21. [29]

    InProceedings of the Nineteenth ACM Conference on Recommender Systems

    Paragon: Parameter Generation for Controllable Multi-Task Recommenda- tion. InProceedings of the Nineteenth ACM Conference on Recommender Systems. 370–380

  22. [30]

    Xu Wang, Jiangxia Cao, Zhiyi Fu, Kun Gai, and Guorui Zhou. 2025. Home: Hierarchy of multi-gate experts for multi-task learning at kuaishou. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    2025

  23. [31]

    Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. InProceedings of the 14th ACM conference on recommender systems. 269–278

    2020

  24. [32]

    Shen Wang, Yusheng Huang, Ruochen Yang, Shuang Wen, Pengbo Xu, Jiangxia Cao, Yueyang Liu, Kuo Cai, Chengcheng Guo, Shiyao Wang, et al. 2026. OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommenda- tion.arXiv preprint arXiv:2602.08612(2026)

    2026

  25. [33]

    Boyang Xia, Zhou Yu, Zhiliang Zhu, Hanxiao Sun, Biyun Han, Jun Wang, Run- nan Liu, and Wenwu Ou. 2026. HarmonRank: Ranking-aligned Multi-objective Ensemble for Live-streaming E-commerce Recommendation.arXiv preprint arXiv:2601.02955(2026)

    2026

  26. [34]

    Yifan Wang, Peijie Sun, Min Zhang, Qinglin Jia, Jingjie Li, and Shaoping Ma

  27. [35]

    Yuya Yoshikawa and Yusaku Imai. 2018. A nonparametric delayed feedback model for conversion rate prediction.arXiv preprint arXiv:1802.00255(2018)

    2018 arXiv

  28. [36]

    Dongbo Xi, Zhen Chen, Peng Yan, Yinger Zhang, Yongchun Zhu, Fuzhen Zhuang, and Yu Chen. 2021. Modeling the sequential dependence among audience multi- step conversions with multi-task learning in targeted display advertising. In Proceedings of the 27th ACM SIGKDD Conference on...

    2021

  29. [37]

    Zhe Zhao, Lichan Hong, Li Wei, Jilin Chen, Aniruddh Nath, Shawn Andrews, Aditee Kumthekar, Maheswaran Sathiamoorthy, Xinyang Yi, and Ed Chi. 2019. Recommending what video to watch next: a multitask ranking system. InPro- ceedings of the 13th ACM conference on recommender syste...

    2019

  30. [38]

    Ruobing Xie, Yanlei Liu, Shaoliang Zhang, Rui Wang, Feng Xia, and Leyu Lin

  31. [41]

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. 2020. Gradient surgery for multi-task learning.Advances in neural information processing systems33 (2020), 5824–5836

    2020

  32. [2021]

    InProceedings of the web conference 2021

    Personalized approximate pareto-efficient recommendation. InProceedings of the web conference 2021. 3839–3849

    2021

  33. [2023]

    InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Unbiased delayed feedback label correction for conversion rate prediction. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2456–2466

  34. [2025]

    InProceed- ings of the Nineteenth ACM Conference on Recommender Systems

    Practical Multi-Task Learning for Rare Conversions in Ad Tech. InProceed- ings of the Nineteenth ACM Conference on Recommender Systems. 1042–1045

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.