Pith. sign in

REVIEW 3 major objections 4 minor 39 references

WatchLens: A Configurable Platform for Online Video Recommendation Experiments

T0 review · 3 major / 4 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 WatchLens ties every playback event to the policy that produced it, at recording time, so video-recommendation experiments no longer need post-hoc joins.

desk verdict Solid open-source infrastructure paper, but the printed event schema does not support the central 'ranking position on every event' claim for watch-page-originated events. read the letter →

arxiv 2608.04807 v1 pith:3KDOVBJO submitted 2026-08-05 cs.IR

Deogyong Kim , Dongha Lee This is my paper · ORCID
classification cs.IR
keywords videorecommendersystemsonlineexperimentationuserstudiesopen-sourceplatformplaybackloggingexposure-awaresession-levelanalysis
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 claims that online video recommendation research lacks a platform that combines flexible manipulation of recommendation conditions with video-native playback observation, and that WatchLens fills this gap. WatchLens is an open-source, single-server system in which interfaces, content pools, and recommendation policies are modular and configurable per participant group, with the feed and watch page able to run different policies. Its defining design choice is a standardized logging layer that attaches the applied policy and ranking position to every playback event at recording time, so policy-to-outcome links do not have to be reconstructed after the fact. The paper demonstrates the platform in a short-form video case study that varies only the watch-page policy and observes session-level playback effects.

What carries the argument

The mechanism that carries the argument is the Event Tracker, a wrapper around the feed page, watch page, and video unit that emits a standardized stream of events under a fixed schema. When an event is emitted, the tracker attaches the exposure context: the page, the applied feed policy, the applied watch-page policy, and the ranking position, so each playback event is self-contained with respect to the recommendation condition that produced it. This removes the need for post-hoc joins between exposure logs and playback logs, which the paper argues can misattribute behavior when recommendation lists change between exposure and interaction.

What would settle it

Set up a WatchLens experiment with a known configuration (for example, a feed policy of 'random' and a watch-page policy of 'similarity'), play a video at a known feed position, and inspect the exported event log: if any playback event carries the wrong policy name, the wrong position, or a missing exposure field, the central claim of exposure-aware linkage fails.

Watch

Extended reading notes

Core claim

The central claim is that exposure-aware playback logging with independently configurable feed and watch-page policies makes controlled online video recommendation experiments accessible outside industrial platforms. The paper's contribution is the architecture itself: every event emitted by any interface carries both the behavioral signal (play, pause, seek, end, watch duration, watch ratio) and the exposure context (which policy produced the impression and at what position), and recommendation policies live in an isolated plug-in layer so they can be swapped without touching the frontend or the logging. The case study, which holds interface, feed policy, and content pool constant while varying only the watch-page policy between a diversity-based and a similarity-based recommender, illustrates the intended use: session length, cumulative watch time, and continuation behavior shift in a consistent direction, and the transition distribution from watch page to feed versus next video is directly readable from the logged events.

Load-bearing premise

The platform's core promise stands on the assumption that the Event Tracker genuinely records the correct policy and ranking position for every event; the paper asserts this design in Section 3.3 but provides no unit tests, ground-truth checks, or audit of the implementation, so a silent logging bug would break the exposure-outcome linkage that makes the platform distinctive.

Editorial extensions

If this is right

  • Researchers can run online video recommendation experiments with a single Docker-deployed system instead of building a one-off pipeline for each study.
  • Because the logging schema is fixed across interfaces, results from studies using different UI presets or custom interfaces are comparable without ad-hoc post-processing.
  • Assigning different policies to the feed and the watch page makes it possible to isolate which recommendation surface drives a behavioral effect, as the case study does for the watch page.
  • The recorded exposure context means researchers can derive additional video-native quantities beyond the built-in metrics, since every exported event carries the policy and position fields.
  • Study configurations, content pools, and policies can be re-run in the same deployable environment, supporting reproducible re-analysis.

Reading between the lines

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

  • If exposure-aware logging proves reliable, the platform could support a standard sanity check for user-study platforms: any competitor can be compared by whether its logs preserve the exact policy and position without post-hoc joining.
  • Because ranking position is recorded per event, a natural testable extension is a study that varies position while holding the policy constant, isolating position effects rather than only policy effects.
  • The feed/watch-page separation invites a 2x2 experimental design that varies both policies together to measure interaction effects, which the paper's single-dimensional case study does not cover.
  • The cross-group comparability claim rests on the logging schema being uniform in practice; an audit that injects known synthetic events and verifies the recorded exposure fields would turn a design assertion into a verified property.
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 presents WatchLens, an open-source platform for online video recommendation experiments. The platform's architecture separates user interfaces, content sources, and recommendation policies, allowing feed and watch-page policies to be assigned independently per user group. A standardized logging layer is claimed to attach the recommendation policy and ranking position to every playback event at recording time, so that policy-to-outcome links do not need to be reconstructed post hoc. The authors demonstrate the platform with a within-subject user study of 30 participants that varies only the watch-page policy while holding the feed policy, interface, and content pool constant, analyzing session-level playback metrics. The paper also releases the platform publicly as a single-server deployable system.

Significance. WatchLens addresses a real gap: existing user-study platforms such as Informfully record video playback separately from recommendation exposure, forcing post-hoc joins that can misattribute events when recommendation lists change. The platform's strengths are concrete: it is open-source under an MIT license, packaged for single-server Docker Compose deployment, and its modular design with independent feed/watch policy assignment is a clean separation that directly supports surface-level experiments. The case study is a workflow demonstration rather than a fitted model, so there is no circularity or free-parameter concern. If the exposure-aware logging works as intended, the platform would be a useful shared resource for academic video recommendation research. However, the significance is moderated by a gap between the paper's central logging claim and the schema actually shown, and by the absence of correctness evidence for the logging path.

major comments (3)
  1. [Abstract and §3.3, Listing 1] The central claim that the logging layer attaches 'the recommendation policy and ranking position to every event at recording time' is not supported for watch-page-originated playback. The single event example records position_in_feed, algorithm_feed, and algorithm_watch, but no field for the position of the item within the watch-page recommendation list. Because §3.2 explicitly supports assigning different policies to the feed and watch page, and because the case study in §4 varies only the watch-page policy, any playback that begins from a watch-page recommendation lacks the ranking position promised by the abstract. If the repository's full schema includes such a field (e.g., position_in_watch), it must be shown in the paper; otherwise the claim should be narrowed to 'policy attached to every event, ranking position for feed-originated events'.
  2. [§3.3, Event Tracker validation] The paper does not validate that the Event Tracker records correct exposure context. No unit tests, seeded-policy checks, or audit of the logging path are reported. For a platform whose main value is that policy-to-outcome links are recorded rather than reconstructed, an implementation bug would silently corrupt every downstream analysis. Please add evidence of correctness: for example, a test that injects a known policy and list, plays a video from a known position, and asserts the emitted event carries the expected policy and position, ideally covering list refresh between exposure and playback.
  3. [Table 1 and §4 (Results)] The case study reports p-values only for four metrics, with no confidence intervals, effect sizes, or multiple-comparison correction across the four tests (plus the transition proportions in Figure 3). With n=30 and correlated within-subject metrics, the statement 'All four metrics shifted consistently... with the largest relative change observed for cumulative watch time' overstates confidence. Report confidence intervals and either correct for multiple testing or explicitly label the comparisons as exploratory.
minor comments (4)
  1. [Listing 1] The key 'pos ition_in _feed' contains stray spaces in the rendering; it should be 'position_in_feed'.
  2. [Headings throughout] The platform name is typeset inconsistently as 'W atchLens' with a space in several headings; this appears to be a LaTeX macro issue and should be fixed.
  3. [§3.1] The backend is said to 'expose the data through analysis endpoints', but the paper does not describe these endpoints or the export format; a sentence on the output schema (e.g., CSV/JSON) would help readers assess reproducibility.
  4. [§4 (Setup)] The paper states participants used WatchLens for at least 20 minutes per day over four days, but does not report actual compliance, the total number of sessions, or the total number of events; a brief compliance summary would strengthen the demonstration.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's claims are architectural and empirical, not derived from fitted parameters or self-cited premises.

full rationale

WatchLens is a systems paper: its central contributions are a modular architecture, a standardized logging schema, and a demonstration case study. There is no derivation chain in which a predicted quantity is computed from a fitted parameter, and no equation that is equivalent to its own input by construction. The case study compares two watch-page policies and reports observed medians with a Wilcoxon test; the metrics are defined from the logged events, not predicted from a model, so nothing is statistically forced. The only mild overlap is that the authors evaluate their own platform, which is standard systems-practice and not circular reasoning. The skeptic's observation about Listing 1 omitting a watch-page ranking position is a potential completeness or correctness issue in the demonstrated schema, not a circularity: even if the position field were absent, that would make the platform's claim false for watch-page events, not make a claim equivalent to its inputs. The paper explicitly refers to the full schema in the repository, so the in-paper excerpt cannot by itself establish a circular reduction. No self-citation is load-bearing: the references to prior work concern external datasets and platforms, and the platform's design decisions are argued from requirements, not from the authors' prior results. Score 0 is therefore appropriate.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper introduces a software system, not a natural entity, and the case study uses no fitted numeric constants. The main unverified premises are the correctness of the logging implementation and the adequacy of the small within-subject study design.

assumptions (4)
  • domain assumption The Event Tracker implementation emits the standardized event schema and attaches the correct policy and position to each event at recording time.
    Section 3.3 states this as a design property, but no tests, ground-truth checks, or audit evidence are provided to confirm the implementation.
  • domain assumption The within-subject, counterbalanced design adequately controls for order effects, fatigue, and individual differences in the case study.
    Section 4 describes balancing on baseline viewing time and reversing policy order across days, but no statistical check for carryover or period effects is reported.
  • domain assumption Thirty university students recruited for the study provide sufficient statistical power for the reported Wilcoxon tests.
    Section 4 reports p-values below 0.05 for all four metrics without a power analysis or effect-size calculation.
  • domain assumption The content pool of 1,000 short-form videos across five categories is suitable for the demonstration and representative enough for the conclusions drawn.
    Section 4 describes the pool but gives no content analysis, external validity check, or comparison with real short-form video distributions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of WatchLens: A Configurable Platform for Online Video Recommendation Experiments." pith.science (2026). https://pith.science/paper/3KDOVBJO

@misc{pith2026260804807,
  author       = {Pith},
  title        = {Pith review of: WatchLens: A Configurable Platform for Online Video Recommendation Experiments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3KDOVBJO}},
  note         = {Machine review of arXiv:2608.04807}
}
read the original abstract

Studying how video recommender systems shape user behavior requires online experiments that link playback behavior with the recommendation conditions that produced it. Existing user-study infrastructure provides one or the other, but not both within a single experimentation workflow. We present WatchLens, an open-source platform that fills this gap. WatchLens adopts a modular architecture in which user interfaces, content sources, and recommendation policies are independently configurable, with policies assignable separately to the feed and the watch page, while a standardized logging layer attaches the recommendation policy and ranking position to every event at recording time. This design enables researchers to analyze how recommendation policies and ranking positions shape downstream playback behavior, session continuation, and navigation between the feed and the watch page, with the linkage between policy and outcome available in each event rather than reconstructed afterwards. We demonstrate WatchLens with a short-form video case study that holds the interface, feed policy, and content pool constant while varying only the watch-page policy, showing how the platform supports session-level comparison of recommendation effects on real viewing behavior. WatchLens is released as a publicly available, single-server deployable system for reproducible online video recommendation research.

Figures

Figures reproduced from arXiv: 2608.04807 by the authors.

Figure 1
Figure 1. WatchLens platform overview. Each user interface is wrapped by an Event Tracker that emits a standardized event stream with exposure context. Recommendation poli￾cies are dispatched through the Plug-in Layer, allowing the interface, feed policy, and watch-page policy to be configured independently per UserGroup over a shared content pool. for rapidly deploying web-based studies on static items such as movies and boo… view at source ↗
Figure 2
Figure 2. WatchLens. (A) and (B) show the researcher-facing admin console: (A) the Groups tab for configuring per-group policies, and (B) the Stats tab with automatically computed metrics. (C) and (D) show the participant-facing interface using the TikTok preset: (C) the feed page and (D) the watch page with related video recommendations beside the playing video. 3.2 Modular Experiment Configuration WatchLens defines a user s… view at source ↗
Figure 3
Figure 3. Distribution of user transitions upon leaving a [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 9 canonical work pages

  1. [1]

    Robin Burke and Michael Ekstrand. 2025. Conducting Recommender Systems User Studies Using POPROX. InAdjunct Proceedings of the 33rd ACM Conference on User Modeling, Adaptation and Personalization (UMAP Adjunct ’25). Association for Computing Machinery, New York, NY, USA, 1–2. doi:10.1145/3708319.3727558

    work page arXiv 2025

  2. [2]

    Qingpeng Cai, Shuchang Liu, Xueliang Wang, Tianyou Zuo, Wentao Xie, Bin Yang, Dong Zheng, Peng Jiang, and Kun Gai. 2023. Reinforcing User Retention in a Billion Scale Short Video Recommender System. InCompanion Proceedings of the ACM Web Conference 2023(Austin, TX, USA)(WWW ’23 Companion). Association for Computing Machinery, New York, NY, USA, 421–426. d...

    arXiv 2023

  3. [3]

    Qingpeng Cai, Zhenghai Xue, Chi Zhang, Wanqi Xue, Shuchang Liu, Ruohan Zhan, Xueliang Wang, Tianyou Zuo, Wentao Xie, Dong Zheng, Peng Jiang, and Kun Gai. 2023. Two-Stage Constrained Actor-Critic for Short Video Recom- mendation. InProceedings of the ACM Web Conference 2023(Austin, TX, USA) (WWW ’23). Association for Computing Machinery, New York, NY, USA,...

    arXiv 2023

  4. [4]

    Minmin Chen, Alex Beutel, Paul Covington, Sagar Jain, Francois Belletti, and Ed H. Chi. 2019. Top-K Off-Policy Correction for a REINFORCE Recommender System. InProceedings of the Twelfth ACM International Conference on Web Search and Data Mining(Melbourne VIC, Australia)(WSDM ’19). Association for Computing Machinery, New York, NY, USA, 456–464. doi:10.11...

    arXiv 2019

  5. [5]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep Neural Networks for YouTube Recommendations. InProceedings of the 10th ACM Conference on Recommender Systems(Boston, Massachusetts, USA)(RecSys ’16). Association for Computing Machinery, New York, NY, USA, 191–198. doi:10.1145/2959100. 2959190

    doi:10.1145/2959100 2016

  6. [6]

    Yashar Deldjoo, Dietmar Jannach, Alejandro Bellogin, Alessandro Difonzo, and Dario Zanzonelli. 2023. Fairness in recommender systems: research landscape and future directions.User Modeling and User-Adapted Interaction34, 1 (April 2023), 59–108. doi:10.1007/s11257-023-09364-z

    doi:10.1007/s11257-023-09364-z 2023

  7. [7]

    Patrik Dokoupil and Ladislav Peska. 2023. EasyStudy: Framework for Easy Deployment of User Studies on Recommender Systems. InProceedings of the 17th ACM Conference on Recommender Systems(Singapore, Singapore)(RecSys ’23). Association for Computing Machinery, New York, NY, USA, 1196–1199. doi:10.1145/3604915.3610640

    arXiv 2023

  8. [8]

    Chongming Gao, Shijun Li, Wenqiang Lei, Jiawei Chen, Biao Li, Peng Jiang, Xiangnan He, Jiaxin Mao, and Tat-Seng Chua. 2022. KuaiRec: A Fully-observed Dataset and Insights for Evaluating Recommender Systems. InProceedings of the 31st ACM International Conference on Information & Knowledge Management (Atlanta, GA, USA)(CIKM ’22). Association for Computing M...

    arXiv 2022

Show all 39 references
  1. [10]

    Florent Garcin, Boi Faltings, Olivier Donatsch, Ayar Alazzawi, Christophe Bruttin, and Amr Huber. 2014. Offline and online evaluation of news recommender systems at swissinfo.ch. InProceedings of the 8th ACM Conference on Recommender Systems(Foster City, Silicon Valley, Califo...

    2014 doi

  2. [11]

    Lucien Heitz, Julian Andrea Croci, Madhav Sachdeva, and Abraham Bernstein

  3. [12]

    Lischka, Alena Birrer, Bibek Paudel, Suzanne Tolmeijer, Laura Laugwitz, and Abraham Bernstein

    Lucien Heitz, Juliane A. Lischka, Alena Birrer, Bibek Paudel, Suzanne Tolmeijer, Laura Laugwitz, and Abraham Bernstein. 2022. Benefits of Diverse News Rec- ommendations for Democracy: A User Study.Digital Journalism10, 10 (2022), 1710–1730. doi:10.1080/21670811.2021.2021804

    2022 arXiv

  4. [13]

    Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk

  5. [14]

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

    2019 doi

  6. [15]

    Yucheng Jin, Nava Tintarev, and Katrien Verbert. 2018. Effects of Personal Charac- teristics on Music Recommender Systems with Different Levels of Controllability. InProceedings of the 12th ACM Conference on Recommender Systems(Vancouver, British Columbia, Canada)(RecSys ’18)....

    2018

  7. [16]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Recom- mendation. In2018 IEEE International Conference on Data Mining (ICDM). IEEE Computer Society, Los Alamitos, CA, USA, 197–206. doi:10.1109/ICDM.2018. 00035

    2018 doi

  8. [17]

    Xiao Lin, Xiaokai Chen, Linfeng Song, Jingwei Liu, Biao Li, and Peng Jiang. 2023. Tree based Progressive Regression Model for Watch-Time Prediction in Short- video Recommendation. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Long Beach...

    2023

  9. [18]

    Chi Liu, Jiangxia Cao, Rui Huang, Kai Zheng, Qiang Luo, Kun Gai, and Guorui Zhou. 2024. KuaiFormer: Transformer-Based Retrieval at Kuaishou.CoRR abs/2411.10057 (2024). https://doi.org/10.48550/arXiv.2411.10057

    2024 doi arXiv

  10. [19]

    Qi Liu, Kai Zheng, Rui Huang, Wuchao Li, Kuo Cai, Yuan Chai, Yanan Niu, Yiqun Hui, Bing Han, Na Mou, Hongning Wang, Wentian Bao, Enyun Yu, Guorui Zhou, Han Li, Yang Song, Defu Lian, and Kun Gai. 2025. RecFlow: An Industrial Full Flow Recommendation Dataset. InThe Thirteenth In...

    2025

  11. [20]

    Zhuoran Liu, Leqi Zou, Xuan Zou, Caihua Wang, Biao Zhang, Da Tang, Bolin Zhu, Yijie Zhu, Peng Wu, Ke Wang, and Youlong Cheng. 2022. Monolith: Real Time Recommendation System with Collisionless Embedding Table. InProceedings of the 5th Workshop on Online Recommender Systems and...

    2022

  12. [21]

    Yunzhu Pan, Chen Gao, Jianxin Chang, Yanan Niu, Yang Song, Kun Gai, De- peng Jin, and Yong Li. 2023. Understanding and Modeling Passive-Negative Feedback for Short-video Sequential Recommendation. InProceedings of the 17th ACM Conference on Recommender Systems(Singapore, Singa...

    2023

  13. [22]

    Aleksandr Poslavsky, Alexander D’yakonov, Yuriy Dorn, and Andrey Zimovnov

  14. [23]

    Tobias Schnabel, Adith Swaminathan, Ashudeep Singh, Navin Chandak, and Thorsten Joachims. 2016. Recommendations as Treatments: Debiasing Learning and Evaluation. InProceedings of the 33rd International Conference on International Conference on Machine Learning - Volume 48(New ...

    2016

  15. [24]

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang

  16. [25]

    Kathrin Wardatzky, Oana Inel, Luca Rossetto, and Abraham Bernstein. 2025. Whom do Explanations Serve? A Systematic Literature Survey of User Charac- teristics in Explainable Recommender Systems Evaluation.ACM Trans. Recomm. Syst.3, 4, Article 49 (April 2025), 35 pages. doi:10....

    2025 doi

  17. [26]

    Wanqi Xue, Qingpeng Cai, Ruohan Zhan, Dong Zheng, Peng Jiang, Kun Gai, and Bo An. 2023. ResAct: Reinforcing Long-term Engagement in Sequential Recommendation with Residual Actor. InThe Eleventh International Conference on Learning Representations. https://openreview.net/forum?...

    2023

  18. [27]

    Xing Yi, Liangjie Hong, Erheng Zhong, Nanthan Nan Liu, and Suju Rajan. 2014. Beyond clicks: dwell time for personalization. InProceedings of the 8th ACM Conference on Recommender Systems(Foster City, Silicon Valley, California, USA) (RecSys ’14). Association for Computing Mach...

    2014

  19. [28]

    Guanghu Yuan, Fajie Yuan, Yudong Li, Beibei Kong, Shujie Li, Lei Chen, Min Yang, Chenyun Yu, Bo Hu, Zang Li, Yu Xu, and Xiaohu Qie. 2022. Tenrec: A Large- scale Multipurpose Benchmark Dataset for Recommender Systems. InAdvances in Neural Information Processing Systems, S. Koye...

    2022 doi

  20. [29]

    Eva Zangerle and Christine Bauer. 2022. Evaluating Recommender Systems: Survey and Framework.ACM Comput. Surv.55, 8, Article 170 (Dec. 2022), 38 pages. doi:10.1145/3556536

    2022 doi

  21. [30]

    Ruohan Zhan, Changhua Pei, Qiang Su, Jianfeng Wen, Xueliang Wang, Guanyu Mu, Dong Zheng, Peng Jiang, and Kun Gai. 2022. Deconfounding Duration Bias in Watch-time Prediction for Video Recommendation. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Dat...

    2022

  22. [31]

    Haiyuan Zhao, Guohao Cai, Jieming Zhu, Zhenhua Dong, Jun Xu, and Ji- Rong Wen. 2024. Counteracting Duration Bias in Video Recommendation RecSys ’26, September 27-October 02, 2026, Minneapolis, MN, USA Kim and Lee via Counterfactual Watch Time. InProceedings of the 30th ACM SIG...

    2024

  23. [32]

    Haiyuan Zhao, Lei Zhang, Jun Xu, Guohao Cai, Zhenhua Dong, and Ji-Rong Wen. 2023. Uncovering User Interest from Biased and Noised Watch Time in Video Recommendation. InProceedings of the 17th ACM Conference on Recom- mender Systems(Singapore, Singapore)(RecSys ’23). Associatio...

    2023

  24. [33]

    Wayne Xin Zhao, Yupeng Hou, Xingyu Pan, Chen Yang, Zeyu Zhang, Zihan Lin, Jingsen Zhang, Shuqing Bian, Jiakai Tang, Wenqi Sun, Yushuo Chen, Lanling Xu, Gaowei Zhang, Zhen Tian, Changxin Tian, Shanlei Mu, Xinyan Fan, Xu Chen, and Ji-Rong Wen. 2022. RecBole 2.0: Towards a More U...

    2022

  25. [34]

    Wayne Xin Zhao, Shanlei Mu, Yupeng Hou, Zihan Lin, Yushuo Chen, Xingyu Pan, Kaiyuan Li, Yujie Lu, Hui Wang, Changxin Tian, Yingqian Min, Zhichao Feng, Xinyan Fan, Xu Chen, Pengfei Wang, Wendi Ji, Yaliang Li, Xiaoling Wang, and Ji-Rong Wen. 2021. RecBole: Towards a Unified, Com...

    2021

  26. [35]

    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

  27. [36]

    Yu Zheng, Chen Gao, Jingtao Ding, Lingling Yi, Depeng Jin, Yong Li, and Meng Wang. 2022. DVR: Micro-Video Recommendation Optimizing Watch-Time-Gain under Duration Bias. InProceedings of the 30th ACM International Conference on Multimedia(Lisboa, Portugal)(MM ’22). Association ...

    2022

  28. [2016]

    In4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, Yoshua Bengio and Yann LeCun (Eds.)

    Session-based Recommendations with Recurrent Neural Networks. In4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, Yoshua Bengio and Yann LeCun (Eds.). http://arxiv.org/abs/1511.06939

    2016 arXiv

  29. [2019]

    InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19)

    BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19). Association for Computing Machinery, New York, NY, USA, ...

  30. [2024]

    InPro- ceedings of the 18th ACM Conference on Recommender Systems(Bari, Italy)(Rec- Sys ’24)

    Informfully - Research Platform for Reproducible User Studies. InPro- ceedings of the 18th ACM Conference on Recommender Systems(Bari, Italy)(Rec- Sys ’24). Association for Computing Machinery, New York, NY, USA, 660–669. doi:10.1145/3640457.3688066

  31. [2026]

    InProceedings of the ACM Web Conference 2026(United Arab Emi- rates)(WWW ’26)

    VK-LSVD: A Large-Scale Industrial Dataset for Short-Video Recom- mendation. InProceedings of the ACM Web Conference 2026(United Arab Emi- rates)(WWW ’26). Association for Computing Machinery, New York, NY, USA, 8657–8660. doi:10.1145/3774904.3792933

    2026

Pith tools

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