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

arXiv:2604.15190v1 [cs.AI] 16 Apr 2026

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

Renbing Chen

Daowei Li

LongCat Interaction Team, Meituan Inc. Beijing, China chenziyang319@163.com

LongCat Interaction Team, Meituan Inc. Beijing, China chenrb@meituan.com

LongCat Interaction Team, Meituan Inc. Beijing, China lidaowei@meituan.com

Jinzhi Liao

Jiashen Sun Ke Zeng

Xiang Zhao∗

Independent Researcher Changsha, China jinzhiliao19@163.com

sunjiashen@meituan.com zengke02@meituan.com LongCat Interaction Team, Meituan Inc. Beijing, China

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.

CCS Concepts • Information systems → Information systems applications.

Keywords Merchant Business Diagnosis, User Behavior Simulation, Large Language Models ∗ Corresponding author.

This work is licensed under a Creative Commons Attribution-NonCommercialNoDerivatives 4.0 International License. SIGIR ’26, Melbourne, VIC, Australia. © 2026 Copyright held by the owner/author(s). ACM ISBN 979-8-4007-2599-9/2026/07 https://doi.org/10.1145/3805712.3808505

Independent Researcher Changsha, China xiangz@aliyun.com

ACM Reference Format: Ziyang Chen, Renbing Chen, Daowei Li, Jinzhi Liao, Jiashen Sun, Ke Zeng, and Xiang Zhao. 2026. Meituan Merchant Business Diagnosis via PolicyGuided Dual-Process User Simulation. In Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’26), July 20–24, 2026, Melbourne, VIC, Australia. ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3805712.3808505

1

Introduction

Merchant business diagnosis on large-scale platforms requires answering counterfactual questions about operational strategies [6, 11, 18, 19]. For example, platform operators need to estimate whether adjusting pricing, restructuring menus, or redesigning promotional pages would improve a merchant’s conversion rate before committing to costly changes. Since each merchant serves a distinct user population with heterogeneous preferences, such diagnosis demands accurate simulation of group-level user behavior under hypothetical interventions. Online A/B testing remains the gold standard for causal evaluation [22, 23], but it suffers from critical scalability limitations. Deploying unproven strategies risks user experience degradation and financial loss [14, 20]. Modern strategies also involve high-dimensional semantic variables such as promotional copywriting and visual layouts [7, 10], creating a combinatorial space that physical experiments cannot cover [3]. These constraints motivate simulation-based counterfactual evaluation [14]. While individual choices are stochastic, the collective preferences of homogeneous user segments exhibit stable regularities, making group-level simulation a natural granularity for merchant diagnosis. Building a trustworthy group-level simulator faces two structural challenges. The first is information incompleteness. Real decisions are influenced by unobserved factors such as offline context and implicit habits, yet reasoning-based simulators rely on explicit inputs and tend to produce over-rationalized predictions [1, 21]. User simulation has evolved from rule-based and statistical approaches [2, 8] to LLM-driven agents [9, 24], but these agents often behave as hyperrational decision-makers that fail to replicate bounded rationality.

SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia.

Ziyang Chen et al.

Figure 1: Overview of the PGHS framework. Phase A abstracts decision policies through macro-level persona clustering and micro-level rationale refinement. Phase B produces predictions via semantic reasoning and statistical fitting, both conditioned on the abstracted policies. Phase C aggregates dual-process outputs into a group-level estimate via population-aligned sampling. The second challenge is mechanism duality. Group behavior simultaneously reflects interpretable preference structures and implicit statistical regularities [12], and a single paradigm typically excels at only one end of this spectrum. Fitting-based simulators built on representation learning [5, 13, 25] capture implicit population-level regularities but degrade under long-tail sparsity and semantic shifts. Despite their complementary strengths, effectively bridging the reasoning and fitting paradigms remains an unexplored challenge. Naive ensembling falls short because the two operate with fundamentally different inductive biases, and unconstrained fusion can easily introduce drift without a shared mechanism to align them. We propose Policy-Guided Hybrid Simulation (PGHS), a dualprocess framework for robust group-level counterfactual evaluation. PGHS mines transferable decision policies from behavioral trajectories to form an explicit mechanism alignment layer, grounding both branches with domain-specific empirical logic [15, 17]. This layer anchors two complementary branches inspired by dual-process cognition [12]. A semantic reasoning branch uses policies to constrain LLM inference and prevent over-rationalization. A statistical fitting branch uses policies as conditional priors to absorb implicit regularities. Predictions from both branches are fused under the shared policy space for complementary correction. Our contributions are threefold. • We identify information incompleteness and mechanism duality as two structural barriers to trustworthy group-level user simulation for merchant diagnosis. • We propose mining decision policies from behavioral trajectories as a shared alignment layer to jointly guide LLM reasoning and statistical fitting. • We deploy PGHS on Meituan with 101 merchants and 26k+ trajectories, achieving 8.80% group simulation error with

45.8% and 40.9% improvements over the best reasoning-based and fitting-based baselines.

2

Method

Given a merchant scene 𝑆, an intervention 𝑎 (e.g., a modified price tier or promotional copy), and a user 𝑢, let 𝑦 (𝑢, 𝑆, 𝑎) ∈ {0, 1} denote the binary purchase outcome. Our goal is to predict the group-level purchase rate E𝑢 [𝑦(𝑢, 𝑆, 𝑎)]. As shown in Figure 1, PGHS operates in three phases: (A) policy mining, (B) dual-branch simulation, and (C) group-level fusion.

2.1

Decision Policy Abstraction

This phase distills noisy online behavioral logs into a concise set of transferable decision policies that serve as explicit proxies for the latent mechanism 𝑢 mech . Trajectory Selection. We first construct a high-intent dataset Dchoice = prof 𝑁 by filtering raw interaction data to retain only {(𝑢𝑖 , 𝑆𝑖 , 𝑦𝑖 )}𝑖=1 sessions where a user actively compares multiple merchants before making a definitive purchase or exit. Hierarchical Policy Discovery. We employ a coarse-to-fine pipeline. At the macro level, we serialize each user profile 𝑢 prof into natural language and map it to a dense embedding hprof = Ψenc (𝑢 prof ). K-Means clustering [16] on these embeddings decomposes the population into 𝐾 persona groups P = {𝑃1, . . . , 𝑃𝐾 }. Within each persona 𝑃𝑘 , we leverage an LLM to generate a textual rationale 𝑟 for each high-intent choice instance (𝑢, 𝑆, 𝑦): 𝑟 = MLLM (Explain 𝑦 | 𝑆, 𝑢 prof ).

(1)

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

We then embed all rationales into a semantic vector space v𝑖rationale = Ψenc (𝑟𝑖 ) and apply HDBSCAN [4] for micro-level refinement: {𝑅𝑘,1, . . . , 𝑅𝑘,𝐽𝑘 }, N𝑘 = HDBSCAN({v𝑖rationale | 𝑖 ∈ 𝑃𝑘 }),

SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia.

)DVW)RRG 6QDFNV

7DLO0HUFKDQWV



&KLQHVH)LQH'LQLQJ







(2)

0LGWLHU0HUFKDQWV

 ([RWLF&XLVLQH

where {𝑅𝑘,𝑗 } are coherent decision sub-groups and N𝑘 are stochastic outliers that we discard.  %DUEHFXH

Dual-Modal Policy Representation. Each refined cluster 𝑅𝑘,𝑗 is formalized into a Decision Policy with two complementary representations. An explicit instruction 𝑇𝑘,𝑗 is generated by prompting an LLM to summarize the shared decision logic: 𝑇𝑘,𝑗 = MLLM (Summarize {𝑟 ∈ 𝑅𝑘,𝑗 }).

(3)

This text contains structured fields including user characteristics, key decision factors, and a decision guide. A dense semantic vector 𝑣𝑘,𝑗 = Ψenc (𝑇𝑘,𝑗 ) is obtained by encoding 𝑇𝑘,𝑗 with the same pre-trained encoder. The explicit instruction guides the reasoning branch, while the dense vector conditions the statistical fitting branch, ensuring both processes operate in a shared semantic space.

2.2

Policy-Guided Dual-Process Simulation

We deploy two parallel branches, both anchored by the shared policy representations: a semantic reasoning branch that leverages explicit decision logic via an LLM, and a statistical fitting branch that captures implicit regularities via a supervised model. Semantic Reasoning. Given a scene 𝑆 and a sampled policy instruction 𝑇𝑘,𝑗 , an LLM-based agent predicts the binary decision: 𝑦ˆreason = 𝑓LLM (𝑆 | SystemPrompt = 𝑇𝑘,𝑗 ).

(4)

The policy text grounds LLM inference in domain-specific trade-off logic, preventing over-rationalization. Statistical Fitting. A supervised ML model (i.e., Gradient Boosting) captures implicit environmental regularities such as geographic convenience and offline brand perception. Unlike standard models, ˜ ours is policy-conditioned. We concatenate scene features 𝜙 (𝑆), user profile embedding eprof , and the policy vector 𝑣𝑘,𝑗 :
˜ eprof, 𝑣𝑘,𝑗 ) . 𝑦ˆfit = 𝑓ML Concat(𝜙 (𝑆), (5) Injecting the policy vector enables the model to learn 𝑃 (𝑌 | 𝑆, Policy), capturing interactions between decision mechanisms and scene features that the text-based reasoning branch cannot observe.

2.3

Group-Level Aggregation

We estimate the policy distribution 𝑃ˆ𝑈 (𝑢 | 𝑆) for each scene using its historical visitor log. Each visitor is assigned to the most similar policy via cosine similarity between their embedding and the policy vectors {𝑣𝑘,𝑗 }. We then conduct Monte Carlo simulation with 𝑁 samples. For each step 𝑖, we draw a policy 𝑢 (𝑖 ) ∼ 𝑃ˆ𝑈 (𝑢 | 𝑆) and obtain predictions from both branches. The final group-level estimate fuses the two: 𝑁 𝑁 1 ∑︁ (𝑖 ) 1 ∑︁ (𝑖 ) 𝑌ˆhybrid (𝑆, 𝑎) = 𝜆 · 𝑦ˆreason + (1 − 𝜆) · 𝑦ˆ , 𝑁 𝑖=1 𝑁 𝑖=1 fit

(6)

where 𝜆 ∈ [0, 1] balances the two branches. This fusion exploits complementary error structures. The reasoning branch generalizes better under data sparsity while the fitting branch absorbs implicit regularities. Empirically, balanced fusion yields the lowest error.



 +RW3RW

D'LVWULEXWLRQE\0HUFKDQW&DWHJRU\

+HDG0HUFKDQWV

E'LVWULEXWLRQE\7UDIILF9ROXPH

Figure 2: Statistical distribution of the dataset by merchant category (a) and traffic volume tier (b).

3 Experiments 3.1 Experimental Setup Datasets. We evaluate PGHS on real-world production data from the Meituan platform, sampled from online user interaction logs during October to November 2025. The evaluation set covers 26,461 user decision trajectories across 101 merchants, spanning five culinary categories and three traffic tiers. Each trajectory records a complete search-to-purchase sequence with explicit comparison actions. User profiles include demographics, consumption habits, and taste preferences. Merchant profiles include category, price tier, rating, and promotional status. All profiles are encoded into dense embeddings via a pre-trained encoder. We adopt a chronological split where the first 20% of trajectories are used for policy mining, and the remaining 80% serve as the test set. Figure 2 shows the distribution of these trajectories across different merchant categories and traffic volume tiers. Baselines. We compare PGHS against two categories of baselines. Fitting-based (ML) baselines predict binary purchase decisions from concatenated user and merchant embeddings. We evaluate Logistic Regression, XGBoost, Gradient Boosting, and a 3-layer DNN. Reasoning-based (LLM) baselines predict purchase decisions from textual descriptions. We evaluate GPT-4o, GPT-4.1, DeepSeek-V3, and Qwen3-32B in both Zero-shot and Few-shot settings. Metrics. We evaluate simulation fidelity using Group Simulation Error (GSE), defined as the mean absolute error between predicted and ground-truth per-merchant choice rates. We also report GSE Standard Deviation (GSE-SD) to measure prediction stability across heterogeneous merchant types. Lower values indicate better accuracy and robustness.

3.2

Overall Simulation Performance

Table 1 reports GSE across merchant tiers and categories. Among ML baselines, DNN (Embedding) achieves the lowest error (14.90%) while Gradient Boosting exhibits the most stable cross-tier performance, though all ML models degrade on Tail merchants due to data sparsity. Among LLM baselines, GPT-4.1 Few-shot achieves the best error (16.25%), but all LLMs exhibit higher variance than ML models due to over-rationalization on implicit-dominant categories such as Snack. PGHS achieves a GSE of 8.80% with a standard deviation of 6.62%, representing relative improvements of 45.8% over the best LLM baseline and 40.9% over the best ML baseline.

SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia.

Ziyang Chen et al.

Table 1: Baseline Comparison: Group Simulation Error (GSE%) across merchant tiers and categories. ML models achieve lower variance but struggle with semantic generalization; LLMs exhibit higher variance due to over-rationality.

Model

Head

Mid

Tail

Chinese

Snack

Exotic

Hotpot

BBQ

GSE (%)

Std

Machine Learning

Logistic Regression XGBoost DNN (Embedding) Gradient Boosting

16.20 14.80 14.20 14.50

17.80 14.97 15.10 14.80

19.50 15.86 16.80 15.60

18.20 16.95 15.80 16.70

17.50 15.06 14.20 14.80

15.80 11.10 12.50 10.80

17.60 16.27 15.30 16.00

16.90 15.11 13.80 14.90

17.80 15.21 14.90 14.98

11.20 9.50 9.80 9.20

Zero-shot LLM

Qwen3-32B DeepSeek-V3 GPT-4o GPT-4.1

15.80 15.20 14.75 14.20

21.10 20.45 20.67 19.80

20.15 19.65 19.41 18.90

21.45 20.90 20.79 20.10

21.30 20.75 20.99 19.50

17.55 16.90 16.61 16.20

18.20 17.55 17.28 16.80

14.65 14.05 13.76 13.20

18.95 18.42 18.27 17.65

13.20 12.85 13.03 12.50

Qwen3-32B DeepSeek-V3 GPT-4o GPT-4.1

14.65 13.95 13.80 13.35

19.20 18.55 18.20 17.65

18.85 18.15 17.80 17.20

19.80 19.20 18.90 18.35

18.65 17.95 17.80 17.25

16.45 15.80 15.50 15.05

17.10 16.50 16.20 15.75

13.95 13.55 13.50 12.95

17.55 16.98 16.80 16.25

12.35 11.95 11.80 11.45

Ours

PGHS (w/o Policy) PGHS (Full)

8.32 7.57

10.43 7.99

11.83 12.26

10.13 8.87

10.16 10.29

10.44 8.22

9.68 8.29

9.16 8.24

9.91 8.80

7.37 6.62

10

(b) Simulation Stability

-15.4%

18.3 15.4

-9.7%

15.2

13.7

-11.2%

9.9

8.8

5 0

LLM LLM ML ML Fusion PGHS Direct +Policy Direct +Policy (No Policy) (Full)

GSE-SD (%)

GSE (%)

15

15.0 12.5 10.0 7.5 5.0 2.5 0.0

13.0

Table 2: Kendall’s 𝜏 between model-predicted and expert strategy rankings, averaged over 10 merchants.

11.8 9.5

8.8

7.4

Model

6.6

XGBoost Gradient Boosting GPT-4.1 (Few-shot) DeepSeek-V3 (Few-shot) PGHS (w/o Policy) PGHS (Full)

LLM LLM ML ML Fusion PGHS Direct +Policy Direct +Policy (No Policy) (Full)

Figure 3: Ablation study on policy guidance and dual-process fusion.

3.3

Ablation Study

Figure 3 compares GSE and GSE-SD across six model variants to isolate the contributions of policy guidance and dual-process fusion. Policy guidance reduces GSE by 15.4% for the LLM branch and 9.7% for the ML branch. Dual-process fusion further amplifies the gains. Even without policy guidance, fusion alone already outperforms the best single-branch model. With policy guidance, PGHS (Full) achieves 8.80% by exploiting complementary error structures. We also compare three variants of the policy discovery pipeline. Without clustering, each user is treated as a unique policy, leading to overfitting and a 23% GSE increase. Using K-Means alone captures coarse persona groups but fails to isolate distinct strategies within each group, resulting in a 12% GSE increase. The full hierarchical pipeline (K-Means + HDBSCAN) yields the best results.

3.4

Overall ↓

Category

(a) Group Simulation Error

20

Merchant Categories (GSE %) ↓

Few-shot LLM

Merchant Tiers (GSE %) ↓

Expert Evaluation for Business Diagnosis

PGHS has been developed and piloted within Meituan as a merchant diagnostic tool. To validate its practical effectiveness, we conduct an expert evaluation. We select 10 merchants spanning diverse categories and tiers, and define 8 candidate operational strategies (e.g., discount promotions, combo meal additions, delivery fee adjustments, storefront redesign). Three senior operations

Kendall’s 𝜏 ↑ 0.45 0.50 0.52 0.50 0.62 0.69

experts independently rank the strategies per merchant by expected effectiveness, and the consensus ranking serves as gold standard. Each model simulates the 8 strategies per merchant and ranks them by predicted conversion rate. We measure ranking quality using Kendall’s 𝜏, which quantifies ordinal agreement with expert consensus (𝜏 ∈ [−1, 1], higher is better). As shown in Table 2, PGHS (Full) achieves a Kendall’s 𝜏 of 0.69, outperforming the best ML baseline by 38.0% and the best LLM baseline by 32.7%. This confirms that PGHS produces strategy rankings well-aligned with expert judgment, validating its utility for real-world merchant diagnosis.

4

Conclusion

We presented PGHS, a policy-guided dual-process framework for group-level user simulation. Deployed on Meituan with 101 merchants and 26k+ trajectories, PGHS achieves 8.80% group simulation error, improving over the best reasoning-based and fitting-based baselines by 45.8% and 40.9% respectively. While our evaluation focuses on merchant-level diagnosis, the dual-process architecture naturally extends to item-level recommendation scenarios where group behavior prediction is required. Future work includes adaptive fusion mechanisms, temporal policy dynamics, and validation on public e-commerce benchmarks.

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

Acknowledgments This work was partially supported by National Natural Science Foundation of China (Nos. U25B2047, 62272469, 72301284), and The Science and Technology Innovation Program of Hunan Province (No. 2023RC1007).

Presenter and Company Bio Ziyang Chen is a research intern in the LongCat Interaction Team at Meituan Inc., focusing on social simulation and LLM-based decisionmaking. He is currently pursuing his PhD at the National University of Defense Technology after receiving his bachelor’s degree in Information and Computing Science. Meituan is China’s leading local services platform, connecting hundreds of millions of consumers with merchants across food delivery, hotel and travel booking, in-store dining, and everyday lifestyle services.

References [1] Carlo Adornetto, Adrian Mora, Kai Hu, Leticia Izquierdo Garcia, Parfait AtchadeAdelomou, 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). [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). [3] Guillaume W Basse and Edoardo M Airoldi. 2018. Limitations of design-based causal inference and A/B testing under arbitrary and network interference. Sociological Methodology 48, 1 (2018), 136–151. [4] Ricardo JGB Campello, Davoud Moulavi, and Jörg Sander. 2013. Density-based clustering based on hierarchical density estimates. In Pacific-Asia conference on knowledge discovery and data mining. Springer, 160–172. [5] Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. 2016. Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems. 7–10. [6] 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 Dong, Yaoyou Fan, Yangyi Fang, Yang Gan, Shiguang Guo, Qi He, Chaowen Hu, Binghui Li, Dailin Li, Xiangyu Li, Yan Li, Chengjian Liu, Xiangfeng Liu, Jiahui Lv, Qiao Ma, Jiang Pan, Cong Qin, Chenxing Sun, Wen Sun, Zhonghui Wang, Abudukelimu Wuerkaixi, Xin Yang, Fangyi Yuan, Yawen Zhu, Tianyi Zhai, Jie Zhang, Runlai Zhang, Yao Xu, Yiran Zhao, Yifan Wang, Xunliang Cai, Yangen Hu, Cao Liu, Lu Pan, Xiaoli Wang, Bo Xiao, Wenyuan Yao, Qianlin Zhou, and Benchang Zhu. 2025. Higher Satisfaction, Lower Cost: A Technical Report on How LLMs Revolutionize Meituan’s Intelligent Interaction Systems. CoRR abs/2510.13291 (2025). [7] Frederick Eberhardt. 2007. Causation and intervention. Unpublished doctoral dissertation, Carnegie Mellon University 93 (2007). [8] Gerhard Fischer. 2001. User modeling in human–computer interaction. User modeling and user-adapted interaction 11, 1 (2001), 65–86. [9] Chen Gao, Xiaochong Lan, Nian Li, Yuan Yuan, Jingtao Ding, Zhilun Zhou, Fengli Xu, and Yong Li. 2024. Large language models empowered agent-based

SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia.

modeling and simulation: A survey and perspectives. Humanities and Social Sciences Communications 11, 1 (2024), 1–24. [10] 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 Methodology 25, 2 (2025), 182–196. [11] James M Joyce. 1999. The foundations of causal decision theory. Cambridge University Press. [12] Daniel Kahneman. 2011. Thinking, fast and slow. macmillan. [13] Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 197–206. [14] Akira Kasuga and Ryo Yonetani. 2024. CXSimulator: A User Behavior Simulation using LLM Embeddings for Web-Marketing Campaign Assessment. In Proceedings 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. [15] Zelong Li, Wenyue Hua, Hao Wang, He Zhu, and Yongfeng Zhang. 2024. FormalLLM: Integrating Formal Language and Natural Language for Controllable LLMbased Agents. ArXiv abs/2402.00798 (2024). [16] James MacQueen. 1967. Some methods for classification and analysis of multivariate observations. In Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 1. Oakland, CA, USA, 281–297. [17] Xing ming Guo, Darioush Keivan, Usman Ahmed Syed, Lianhui Qin, Huan Zhang, Geir E. Dullerud, Peter J. Seiler, and Bin Hu. 2024. ControlAgent: Automating Control System Design via Novel Integration of LLM Agents and Domain Expertise. ArXiv abs/2410.19811 (2024). [18] Judea Pearl. 2009. Causal inference in statistics: An overview. (2009). [19] 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 journal 9, 4 (2015), 351–379. [20] Craig Silverstein, Sergey Brin, Rajeev Motwani, and Jeffrey D. Ullman. 1998. Scalable Techniques for Mining Causal Structures. In VLDB’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. [21] Wenjie Wang, Fuli Feng, Xiangnan He, Xiang Wang, and Tat-Seng Chua. 2021. Deconfounded recommendation for alleviating bias amplification. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1717– 1725. [22] Xuan Yin and Liangjie Hong. 2019. The Identification and Estimation of Direct and Indirect Effects in A/B Tests through Causal Mediation Analysis. In Proceedings 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 Terzi, and George Karypis (Eds.). ACM, 2989–2999. [23] Yuan Yuan, Kristen M. Altenburger, and Farshad Kooti. 2021. Causal Network Motifs: Identifying Heterogeneous Spillover Effects in A/B Tests. In WWW ’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. [24] 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. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 13339–13347. [25] 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 clickthrough rate prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 1059–1068.

Related documents

• Intégration énergétique des procédés industriels par la méthode du pincement.#198577
• THE ATTITUDE OF MERCHANTS TOWARDS FATWA ON RIBA AND CONVENTIONAL BANK IN LANGSA#936641
• Training on the Implementation of the Halal Product Assurance System for Micro and Small Enterprises to Increase Their Intention to Obtain Halal Certification#275901
• Training on the Implementation of the Halal Product Assurance System for Micro and Small Enterprises to Increase Their Intention to Obtain Halal Certification#274508
• Pembuatan Alat Bantu Material Handling untuk Menurunkan Tingkat Kelelahan Fisik Operator#678038