Meituan Merchant Business Diagnosis via Policy-Guided Dual-Process User Simulation

Daowei Li; Jiashen Sun; Jinzhi Liao; Ke Zeng; Renbing Chen; Xiang Zhao; Ziyang Chen

arxiv: 2604.15190 · v1 · submitted 2026-04-16 · 💻 cs.AI · cs.CL

Meituan Merchant Business Diagnosis via Policy-Guided Dual-Process User Simulation

Ziyang Chen , Renbing Chen , Daowei Li , Jinzhi Liao , Jiashen Sun , Ke Zeng , Xiang Zhao This is my paper

Pith reviewed 2026-05-10 11:05 UTC · model grok-4.3

classification 💻 cs.AI cs.CL

keywords user behavior simulationdual-process frameworkpolicy-guided alignmentLLM reasoningmachine learning fittingmerchant strategy evaluationbehavioral trajectoriescounterfactual simulation

0 comments

The pith

Policy-Guided Hybrid Simulation fuses LLM reasoning and ML fitting via mined policies to reach 8.80% group behavior error.

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

The paper develops a simulator for group-level user behavior that lets merchants test strategies through counterfactual evaluation instead of running costly online experiments. It targets two problems: reasoning models over-rationalize when data on offline context or habits is missing, and no single method captures both explicit preferences and hidden statistical patterns. The solution mines decision policies from real trajectories to guide an LLM reasoning branch and an ML fitting branch, then fuses the group-level outputs for mutual correction.

Core claim

Policy-Guided Hybrid Simulation (PGHS) mines transferable decision policies from behavioral trajectories to serve as a shared alignment layer. This layer anchors an LLM-based reasoning branch that prevents over-rationalization and an ML-based fitting branch that absorbs implicit regularities, after which group-level predictions from both branches are fused for complementary correction.

What carries the argument

The policy-guided alignment layer that mines decision policies from trajectories to anchor and fuse LLM reasoning and ML fitting branches.

If this is right

Merchants gain a scalable way to evaluate strategies without live experiments.
Group simulation error reaches 8.80%, a 45.8% reduction from the strongest reasoning baseline.
The system captures both interpretable decision rules and statistical regularities in one pipeline.
Fusion of the two branches supplies error correction that neither branch achieves alone.

Where Pith is reading between the lines

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

The same policy-mining step could anchor simulators for user behavior on other platforms that collect trajectory data.
Lower error in group predictions could shorten the cycle of testing and refining merchant tactics in e-commerce.
Checking how well the mined policies transfer when merchant categories or user demographics shift would test the framework's robustness beyond the reported deployment.

Load-bearing premise

Policies mined from observed trajectories are sufficiently transferable to anchor LLM reasoning against over-rationalization and to enable complementary correction when fused with the ML branch.

What would settle it

Running PGHS on a fresh collection of merchants and trajectories where the fused error rate fails to beat the better of the two separate branches would falsify the value of the policy layer and fusion step.

Figures

Figures reproduced from arXiv: 2604.15190 by Daowei Li, Jiashen Sun, Jinzhi Liao, Ke Zeng, Renbing Chen, Xiang Zhao, Ziyang Chen.

**Figure 1.** Figure 1: Overview of the PGHS framework. Phase A abstracts decision policies through macro-level persona clustering and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: Statistical distribution of the dataset by merchant [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Ablation study on policy guidance and dual-process [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

read the original abstract

Simulating group-level user behavior enables scalable counterfactual evaluation of merchant strategies without costly online experiments. However, building a trustworthy simulator faces two structural challenges. First, information incompleteness causes reasoning-based simulators to over-rationalize when unobserved factors such as offline context and implicit habits are missing. Second, mechanism duality requires capturing both interpretable preferences and implicit statistical regularities, which no single paradigm achieves alone. We propose Policy-Guided Hybrid Simulation (PGHS), a dual-process framework that mines transferable decision policies from behavioral trajectories and uses them as a shared alignment layer. This layer anchors an LLM-based reasoning branch that prevents over-rationalization and an ML-based fitting branch that absorbs implicit regularities. Group-level predictions from both branches are fused for complementary correction. We deploy PGHS on Meituan with 101 merchants and over 26,000 trajectories. PGHS achieves a group simulation error of 8.80%, improving over the best reasoning-based and fitting-based baselines by 45.8% and 40.9% respectively.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

PGHS shows a practical hybrid for Meituan user simulation with clear error cuts, but the policy transfer step needs explicit checks to confirm the gains come from the dual-process design.

read the letter

The core takeaway is that this paper mines decision policies from behavioral trajectories and uses them to align an LLM reasoning branch with an ML fitting branch, then fuses the outputs for group-level user simulation on Meituan. It reports an 8.8% group error, beating the best single-paradigm baselines by 40-45% on data from 101 merchants and 26k trajectories. That deployment gives the numbers some grounding in real platform use rather than synthetic tests. The new element is the policy layer acting as an explicit anchor: it constrains the LLM against over-rationalizing unobserved factors like offline habits while letting the ML branch capture statistical patterns the LLM misses. The fusion step for complementary correction is a reasonable way to combine the two. The paper does well by focusing on a concrete e-commerce problem and measuring at the group level instead of individual predictions. The soft spots are mostly around verification. The abstract gives no information on how policies are mined, whether mining and evaluation use disjoint trajectory sets, or how baselines were implemented. If policy extraction draws from the same data later used for testing, the reported gains could partly reflect leakage instead of the anchoring mechanism. No error bars or split details appear either, which makes it hard to judge if the 40%+ improvements are stable. The stress-test note on transferability is fair to raise from the abstract alone; the full paper would need to show cross-merchant or held-out policy performance to close that gap. This work is aimed at applied teams building simulators for merchant strategy testing in recommendation or e-commerce platforms. Readers who already work with hybrid LLM-ML setups or need counterfactuals without live experiments will find the architecture and numbers useful to adapt. It deserves peer review because the real deployment and measurable improvements are substantive enough to warrant checking the methods and data handling in detail.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes Policy-Guided Hybrid Simulation (PGHS), a dual-process framework for group-level user behavior simulation on Meituan merchant data. It mines transferable decision policies from behavioral trajectories to serve as an alignment layer that anchors an LLM-based reasoning branch (to mitigate over-rationalization from missing factors) and an ML-based fitting branch (to capture implicit regularities), then fuses the group-level predictions from both branches. On a dataset of 101 merchants and over 26,000 trajectories, PGHS reports a group simulation error of 8.80%, with relative improvements of 45.8% over the best reasoning-based baseline and 40.9% over the best fitting-based baseline.

Significance. If the transferability and fusion claims hold under proper held-out evaluation, the work offers a practical approach to scalable counterfactual merchant strategy evaluation by explicitly combining interpretable policy guidance with statistical and LLM components. The large-scale real-world deployment on Meituan data and the explicit handling of information incompleteness are notable strengths; the empirical gains, if robust, could reduce reliance on online experiments in e-commerce settings.

major comments (2)

[Experimental evaluation / §5] The central performance claim (8.80% group error and the 45.8%/40.9% relative gains) depends on the assumption that policies mined from observed trajectories are transferable and provide complementary signal without leakage. The manuscript provides no details on the data splitting procedure for policy mining versus evaluation (e.g., whether mining occurs only on training trajectories or includes the held-out set used to compute the reported errors). This is load-bearing for the dual-process anchoring claim and must be clarified with explicit train/test protocol in the experimental section.
[Results / Table reporting the 8.80% error and baseline comparisons] No error bars, standard deviations, or statistical significance tests are reported for the 8.80% error or the relative improvements over baselines. Given that the results are the primary evidence for the framework's superiority, the absence of variability measures prevents assessment of whether the gains are robust or could arise from post-hoc baseline choices or single-run variance.

minor comments (1)

[Abstract] The abstract states the number of merchants and trajectories but does not specify the time span, merchant selection criteria, or trajectory length distribution; adding these would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which helps strengthen the clarity and rigor of our work. We address each major comment below and will update the manuscript accordingly.

read point-by-point responses

Referee: [Experimental evaluation / §5] The central performance claim (8.80% group error and the 45.8%/40.9% relative gains) depends on the assumption that policies mined from observed trajectories are transferable and provide complementary signal without leakage. The manuscript provides no details on the data splitting procedure for policy mining versus evaluation (e.g., whether mining occurs only on training trajectories or includes the held-out set used to compute the reported errors). This is load-bearing for the dual-process anchoring claim and must be clarified with explicit train/test protocol in the experimental section.

Authors: We agree that the data splitting protocol is essential to validate transferability and rule out leakage. In the reported experiments, policy mining was performed exclusively on training trajectories for each of the 101 merchants; held-out test trajectories (the 26k+ used for error computation) were never accessed during mining, alignment, or branch training. We will add an explicit description of the train/test protocol—including split ratios (e.g., per-merchant temporal or random 70/30), confirmation that test data remained unseen during policy extraction, and how this supports the dual-process claims—to the revised §5. revision: yes
Referee: [Results / Table reporting the 8.80% error and baseline comparisons] No error bars, standard deviations, or statistical significance tests are reported for the 8.80% error or the relative improvements over baselines. Given that the results are the primary evidence for the framework's superiority, the absence of variability measures prevents assessment of whether the gains are robust or could arise from post-hoc baseline choices or single-run variance.

Authors: We acknowledge that variability measures and significance tests are needed to demonstrate robustness. We will re-execute the ML fitting branch across multiple random seeds, report standard deviations as error bars on the 8.80% error and all baseline comparisons, and add statistical significance tests (paired t-tests or Wilcoxon signed-rank tests) between PGHS and the best baselines. These updates will appear in the revised results table and accompanying text. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper reports an empirical group simulation error of 8.80% on held-out Meituan trajectories (101 merchants, >26k trajectories), with gains over baselines. Policy mining from observed trajectories is used to anchor LLM and ML branches before fusion, but the performance metric is measured externally on separate data rather than reducing to any fitted parameter or input by construction. No equations, self-definitional steps, or load-bearing self-citations are exhibited that force the result. The evaluation remains independent of the internal mechanisms.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that behavioral trajectories yield transferable policies capable of constraining LLM over-rationalization and enabling branch fusion; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)

domain assumption Behavioral trajectories contain extractable transferable decision policies that can serve as a shared alignment layer for reasoning and fitting branches.
Invoked as the core mechanism that addresses information incompleteness and mechanism duality.

pith-pipeline@v0.9.0 · 5498 in / 1236 out tokens · 52673 ms · 2026-05-10T11:05:08.838120+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · 1 internal anchor

[1]

Carlo Adornetto, Adrian Mora, Kai Hu, Leticia Izquierdo Garcia, Parfait Atchade- Adelomou, Gianluigi Greco, Luis Alberto Alonso Pastor, and Kent Larson. 2025. Generative agents in agent-based modeling: Overview, validation, and emerging challenges.IEEE Transactions on Artificial Intelligence(2025)

work page 2025
[2]

Krisztian Balog and ChengXiang Zhai. 2025. User simulation in the era of generative ai: User modeling, synthetic data generation, and system evaluation. arXiv preprint arXiv:2501.04410(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[3]

Guillaume W Basse and Edoardo M Airoldi. 2018. Limitations of design-based causal inference and A/B testing under arbitrary and network interference.Soci- ological Methodology48, 1 (2018), 136–151

work page 2018
[4]

Ricardo JGB Campello, Davoud Moulavi, and Jörg Sander. 2013. Density-based clustering based on hierarchical density estimates. InPacific-Asia conference on knowledge discovery and data mining. Springer, 160–172

work page 2013
[5]

Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al

work page
[6]

InProceedings of the 1st workshop on deep learning for recommender systems

Wide & deep learning for recommender systems. InProceedings of the 1st workshop on deep learning for recommender systems. 7–10

work page
[7]

Xuxin Cheng, Ke Zeng, Zhiquan Cao, Linyi Dai, Wenxuan Gao, Fei Han, Ai Jian, Feng Hong, Wenxing Hu, Zihe Huang, Dejian Kong, Jia Leng, Zhuoyuan Liao, Pei Liu, Jiaye Lin, Xing Ma, Jingqing Ruan, Jiaxing Song, Xiaoyu Tan, Ruixuan Xiao, Wenhui Yu, Wenyu Zhan, Haoxing Zhang, Chao Zhou, Hao Zhou, Shaodong Zheng, Ruinian Chen, Siyuan Chen, Ziyang Chen, Yiwen Do...

work page arXiv 2025
[8]

Frederick Eberhardt. 2007. Causation and intervention.Unpublished doctoral dissertation, Carnegie Mellon University93 (2007)

work page 2007
[9]

Gerhard Fischer. 2001. User modeling in human–computer interaction.User modeling and user-adapted interaction11, 1 (2001), 65–86

work page 2001
[10]

Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2024. Large language models empowered agent-based modeling and simulation: A survey and perspectives.Humanities and Social Sciences Communications11, 1 (2024), 1–24

work page 2024
[11]

Sarah Ruth Hoffman, Nilesh Gangan, Xiaoxue Chen, Joseph L Smith, Arlene Tave, Yiling Yang, Christopher L Crowe, Susan dosReis, and Michael Grabner. 2025. A step-by-step guide to causal study design using real-world data.Health Services and Outcomes Research Methodology25, 2 (2025), 182–196

work page 2025
[12]

1999.The foundations of causal decision theory

James M Joyce. 1999.The foundations of causal decision theory. Cambridge University Press

work page 1999
[13]

2011.Thinking, fast and slow

Daniel Kahneman. 2011.Thinking, fast and slow. macmillan

work page 2011
[14]

Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206

work page 2018
[15]

Akira Kasuga and Ryo Yonetani. 2024. CXSimulator: A User Behavior Simulation using LLM Embeddings for Web-Marketing Campaign Assessment. InProceed- ings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21-25, 2024, Edoardo Serra and Francesca Spezzano (Eds.). ACM, 3817–3821

work page 2024
[16]

Zelong Li, Wenyue Hua, Hao Wang, He Zhu, and Yongfeng Zhang. 2024. Formal- LLM: Integrating Formal Language and Natural Language for Controllable LLM- based Agents.ArXivabs/2402.00798 (2024)

work page arXiv 2024
[17]

James MacQueen. 1967. Some methods for classification and analysis of multivari- ate observations. InProceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 1. Oakland, CA, USA, 281–297

work page 1967
[18]

ControlAgent: Automating control system design via novel integration of LLM agents and domain expertise,

Xing ming Guo, Darioush Keivan, Usman Ahmed Syed, Lianhui Qin, Huan Zhang, Geir E. Dullerud, Peter J. Seiler, and Bin Hu. 2024. ControlAgent: Automating Con- trol System Design via Novel Integration of LLM Agents and Domain Expertise. ArXivabs/2410.19811 (2024)

work page arXiv 2024
[19]

Judea Pearl. 2009. Causal inference in statistics: An overview. (2009)

work page 2009
[20]

Isabelle MMJ Reymen, Petra Andries, Hans Berends, Rene Mauer, Ute Stephan, and Elco Van Burg. 2015. Understanding dynamics of strategic decision making in venture creation: a process study of effectuation and causation.Strategic entrepreneurship journal9, 4 (2015), 351–379

work page 2015
[21]

Craig Silverstein, Sergey Brin, Rajeev Motwani, and Jeffrey D. Ullman. 1998. Scalable Techniques for Mining Causal Structures. InVLDB’98, Proceedings of 24rd International Conference on Very Large Data Bases, August 24-27, 1998, New York City, New York, USA, Ashish Gupta, Oded Shmueli, and Jennifer Widom (Eds.). Morgan Kaufmann, 594–605

work page 1998
[22]

Wenjie Wang, Fuli Feng, Xiangnan He, Xiang Wang, and Tat-Seng Chua. 2021. Deconfounded recommendation for alleviating bias amplification. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1717– 1725

work page 2021
[23]

Xuan Yin and Liangjie Hong. 2019. The Identification and Estimation of Direct and Indirect Effects in A/B Tests through Causal Mediation Analysis. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2019, Anchorage, AK, USA, August 4-8, 2019, Ankur Teredesai, Vipin Kumar, Ying Li, Rómer Rosales, Evimaria...

work page 2019
[24]

Altenburger, and Farshad Kooti

Yuan Yuan, Kristen M. Altenburger, and Farshad Kooti. 2021. Causal Network Motifs: Identifying Heterogeneous Spillover Effects in A/B Tests. InWWW ’21: The Web Conference 2021, Virtual Event / Ljubljana, Slovenia, April 19-23, 2021, Jure Leskovec, Marko Grobelnik, Marc Najork, Jie Tang, and Leila Zia (Eds.). ACM / IW3C2, 3359–3370

work page 2021
[25]

Zijian Zhang, Shuchang Liu, Ziru Liu, Rui Zhong, Qingpeng Cai, Xiangyu Zhao, Chunxu Zhang, Qidong Liu, and Peng Jiang. 2025. Llm-powered user simulator for recommender system. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 13339–13347

work page 2025
[26]

Guorui Zhou, Xiaoqiang Zhu, Chengru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click- through rate prediction. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1059–1068

work page 2018

[1] [1]

Carlo Adornetto, Adrian Mora, Kai Hu, Leticia Izquierdo Garcia, Parfait Atchade- Adelomou, Gianluigi Greco, Luis Alberto Alonso Pastor, and Kent Larson. 2025. Generative agents in agent-based modeling: Overview, validation, and emerging challenges.IEEE Transactions on Artificial Intelligence(2025)

work page 2025

[2] [2]

Krisztian Balog and ChengXiang Zhai. 2025. User simulation in the era of generative ai: User modeling, synthetic data generation, and system evaluation. arXiv preprint arXiv:2501.04410(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025

[3] [3]

Guillaume W Basse and Edoardo M Airoldi. 2018. Limitations of design-based causal inference and A/B testing under arbitrary and network interference.Soci- ological Methodology48, 1 (2018), 136–151

work page 2018

[4] [4]

Ricardo JGB Campello, Davoud Moulavi, and Jörg Sander. 2013. Density-based clustering based on hierarchical density estimates. InPacific-Asia conference on knowledge discovery and data mining. Springer, 160–172

work page 2013

[5] [5]

Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al

work page

[6] [6]

InProceedings of the 1st workshop on deep learning for recommender systems

Wide & deep learning for recommender systems. InProceedings of the 1st workshop on deep learning for recommender systems. 7–10

work page

[7] [7]

Xuxin Cheng, Ke Zeng, Zhiquan Cao, Linyi Dai, Wenxuan Gao, Fei Han, Ai Jian, Feng Hong, Wenxing Hu, Zihe Huang, Dejian Kong, Jia Leng, Zhuoyuan Liao, Pei Liu, Jiaye Lin, Xing Ma, Jingqing Ruan, Jiaxing Song, Xiaoyu Tan, Ruixuan Xiao, Wenhui Yu, Wenyu Zhan, Haoxing Zhang, Chao Zhou, Hao Zhou, Shaodong Zheng, Ruinian Chen, Siyuan Chen, Ziyang Chen, Yiwen Do...

work page arXiv 2025

[8] [8]

Frederick Eberhardt. 2007. Causation and intervention.Unpublished doctoral dissertation, Carnegie Mellon University93 (2007)

work page 2007

[9] [9]

Gerhard Fischer. 2001. User modeling in human–computer interaction.User modeling and user-adapted interaction11, 1 (2001), 65–86

work page 2001

[10] [10]

Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2024. Large language models empowered agent-based modeling and simulation: A survey and perspectives.Humanities and Social Sciences Communications11, 1 (2024), 1–24

work page 2024

[11] [11]

Sarah Ruth Hoffman, Nilesh Gangan, Xiaoxue Chen, Joseph L Smith, Arlene Tave, Yiling Yang, Christopher L Crowe, Susan dosReis, and Michael Grabner. 2025. A step-by-step guide to causal study design using real-world data.Health Services and Outcomes Research Methodology25, 2 (2025), 182–196

work page 2025

[12] [12]

1999.The foundations of causal decision theory

James M Joyce. 1999.The foundations of causal decision theory. Cambridge University Press

work page 1999

[13] [13]

2011.Thinking, fast and slow

Daniel Kahneman. 2011.Thinking, fast and slow. macmillan

work page 2011

[14] [14]

Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206

work page 2018

[15] [15]

Akira Kasuga and Ryo Yonetani. 2024. CXSimulator: A User Behavior Simulation using LLM Embeddings for Web-Marketing Campaign Assessment. InProceed- ings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21-25, 2024, Edoardo Serra and Francesca Spezzano (Eds.). ACM, 3817–3821

work page 2024

[16] [16]

Zelong Li, Wenyue Hua, Hao Wang, He Zhu, and Yongfeng Zhang. 2024. Formal- LLM: Integrating Formal Language and Natural Language for Controllable LLM- based Agents.ArXivabs/2402.00798 (2024)

work page arXiv 2024

[17] [17]

James MacQueen. 1967. Some methods for classification and analysis of multivari- ate observations. InProceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 1. Oakland, CA, USA, 281–297

work page 1967

[18] [18]

ControlAgent: Automating control system design via novel integration of LLM agents and domain expertise,

Xing ming Guo, Darioush Keivan, Usman Ahmed Syed, Lianhui Qin, Huan Zhang, Geir E. Dullerud, Peter J. Seiler, and Bin Hu. 2024. ControlAgent: Automating Con- trol System Design via Novel Integration of LLM Agents and Domain Expertise. ArXivabs/2410.19811 (2024)

work page arXiv 2024

[19] [19]

Judea Pearl. 2009. Causal inference in statistics: An overview. (2009)

work page 2009

[20] [20]

Isabelle MMJ Reymen, Petra Andries, Hans Berends, Rene Mauer, Ute Stephan, and Elco Van Burg. 2015. Understanding dynamics of strategic decision making in venture creation: a process study of effectuation and causation.Strategic entrepreneurship journal9, 4 (2015), 351–379

work page 2015

[21] [21]

Craig Silverstein, Sergey Brin, Rajeev Motwani, and Jeffrey D. Ullman. 1998. Scalable Techniques for Mining Causal Structures. InVLDB’98, Proceedings of 24rd International Conference on Very Large Data Bases, August 24-27, 1998, New York City, New York, USA, Ashish Gupta, Oded Shmueli, and Jennifer Widom (Eds.). Morgan Kaufmann, 594–605

work page 1998

[22] [22]

Wenjie Wang, Fuli Feng, Xiangnan He, Xiang Wang, and Tat-Seng Chua. 2021. Deconfounded recommendation for alleviating bias amplification. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1717– 1725

work page 2021

[23] [23]

Xuan Yin and Liangjie Hong. 2019. The Identification and Estimation of Direct and Indirect Effects in A/B Tests through Causal Mediation Analysis. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2019, Anchorage, AK, USA, August 4-8, 2019, Ankur Teredesai, Vipin Kumar, Ying Li, Rómer Rosales, Evimaria...

work page 2019

[24] [24]

Altenburger, and Farshad Kooti

Yuan Yuan, Kristen M. Altenburger, and Farshad Kooti. 2021. Causal Network Motifs: Identifying Heterogeneous Spillover Effects in A/B Tests. InWWW ’21: The Web Conference 2021, Virtual Event / Ljubljana, Slovenia, April 19-23, 2021, Jure Leskovec, Marko Grobelnik, Marc Najork, Jie Tang, and Leila Zia (Eds.). ACM / IW3C2, 3359–3370

work page 2021

[25] [25]

Zijian Zhang, Shuchang Liu, Ziru Liu, Rui Zhong, Qingpeng Cai, Xiangyu Zhao, Chunxu Zhang, Qidong Liu, and Peng Jiang. 2025. Llm-powered user simulator for recommender system. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 13339–13347

work page 2025

[26] [26]

Guorui Zhou, Xiaoqiang Zhu, Chengru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click- through rate prediction. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1059–1068

work page 2018