Zuohan Wu

Zuohan Wu

PhD candidate at HKUST(GZ) | Data Science & Analytics

Publications

You can also find my articles on Google Scholar and ORCID.

Selected Publications

Saturn: Agentic Natural Language-Driven Tabular Data Discovery

Accepted in VLDB 2027 (to appear).

Saturn uses a multi-agent framework to construct a user preference-based ontology and reusable table profiles for training-free, natural-language-driven tabular data discovery.

Recommended citation: Zuohan Wu, Jiachuan Wang, Jiake Ge, Xujia Li, Libin Zheng, Yongqi Zhang, Shuangyin Li, Lei Chen. "Saturn: Agentic Natural Language-Driven Tabular Data Discovery." Proceedings of the VLDB Endowment (VLDB 2027), accepted; to appear.

Code

Superficial Reflection or Genuine Thought? A Fine-Grained Cognitive Analysis of Large Reasoning Models

Accepted in Findings of AACL-IJCNLP 2026 (to appear).

A fine-grained taxonomy and the CAPO annotation framework reveal how large reasoning models organise information, reflect, and self-correct, with validation across model generations and reasoning domains.

Recommended citation: Yuxiang Chen*, Zuohan Wu*, Ziwei Wang, Xiangning Yu, Xujia Li, Linyi Yang, Mengyue Yang, Jun Wang, Lei Chen. "Superficial Reflection or Genuine Thought? A Fine-Grained Cognitive Analysis of Large Reasoning Models." Findings of AACL-IJCNLP 2026, accepted. *Equal contribution.

PDF Earlier arXiv version Code

RuleMem: Active Rule Memory for Long-Term Conversational Agents

Preprint, 2026.

RuleMem learns reusable natural-language rules from dialogue history to guide memory retrieval and answer generation in long-term conversational agents.

Recommended citation: Xingyuan Zeng, Zuohan Wu, Yue Wang, Chen Zhang, Quanming Yao, Wei Liu, Jiuke Wang, Libin Zheng, Jian Yin. "RuleMem: Active Rule Memory for Long-Term Conversational Agents." arXiv preprint arXiv:2609.03915, 2026.

PDF arXiv

Efficient Zero-Shot and Label-free Log Anomaly Detection for Resource-Constrained Systems

Published in 2026 IEEE 42nd International Conference on Data Engineering (ICDE).

MaidLog uses LLM-assisted pseudo-labels to train a generalisable, lightweight log anomaly detector. Downstream detection requires neither manual labels nor LLM inference, making it suitable for resource-constrained systems.

Recommended citation: Zuohan Wu, Jiachuan Wang, Libin Zheng, Yongqi Zhang, Shuangyin Li, Lei Chen. "Efficient Zero-Shot and Label-free Log Anomaly Detection for Resource-Constrained Systems." 2026 IEEE 42nd International Conference on Data Engineering (ICDE), pp. 671-684, 2026.

PDF DOI Code DBLP

DA-RAG: Dynamic Attributed Community Search for Retrieval-Augmented Generation

Published in Proceedings of the ACM Web Conference 2026 (WWW 2026).

This paper proposes DA-RAG, which leverages attributed community search to dynamically extract relevant subgraphs for retrieval-augmented generation, outperforming existing RAG methods by up to 40%.

Recommended citation: Xingyuan Zeng, Zuohan Wu, Yue Wang, Chen Zhang, Quanming Yao, Libin Zheng, Jian Yin. "DA-RAG: Dynamic Attributed Community Search for Retrieval-Augmented Generation." Proceedings of the ACM Web Conference 2026, pp. 2195-2206, 2026.

PDF DOI arXiv Code DBLP

VPLight: A Reinforcement Learning Approach for Traffic Signal Control With Pedestrian Dynamics

Published in IEEE Transactions on Knowledge and Data Engineering (TKDE), 2026.

This paper proposes VPLight, a reinforcement learning approach that considers pedestrian dynamics for more effective traffic signal control at intersections.

Recommended citation: Xinyu Zhang, Zuohan Wu, Chen Jason Zhang, Libin Zheng, Peng Cheng, Jian Yin, Cyrus Shahabi. "VPLight: A Reinforcement Learning Approach for Traffic Signal Control With Pedestrian Dynamics." IEEE Transactions on Knowledge and Data Engineering, vol. 38, no. 3, pp. 2079-2093, 2026. https://doi.org/10.1109/TKDE.2025.3641213

DBLP

DRLPG: Reinforced Opponent-Aware Order Pricing for Hub Mobility Services

Published in IEEE Transactions on Knowledge and Data Engineering (TKDE), 2025.

This paper develops a deep reinforcement learning model with a multi-policy fusion mechanism that enables more adaptive and profitable pricing strategies in dynamic competitive environments.

Recommended citation: Zuohan Wu, Chen Jason Zhang, Han Yin, Rui Meng, Libin Zheng, Huaijie Zhu, Wei Liu. "DRLPG: Reinforced Opponent-Aware Order Pricing for Hub Mobility Services." IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 6, pp. 3298-3311, 2025. https://doi.org/10.1109/TKDE.2025.3551147

Opponent-aware Order Pricing towards Hub-oriented Mobility Services

Published in 39th IEEE International Conference on Data Engineering (ICDE 2023).

This paper proposes a novel opponent-aware pricing framework for emerging hub-oriented mobility services, significantly improving platform revenue and operational efficiency.

Recommended citation: Zuohan Wu, Libin Zheng, Chen Jason Zhang, Huaijie Zhu, Jian Yin, Di Jiang. "Opponent-aware Order Pricing towards Hub-oriented Mobility Services." ICDE 2023, pp. 1874-1886. https://doi.org/10.1109/ICDE55515.2023.00146

Full Publication List

Forthcoming

2026

2025

2023

Research Topics

My research spans several interconnected areas:

  1. Large Language Models for Data Engineering: Developing LLM-based systems for automating data engineering tasks, including log analysis and anomaly detection.

  2. Deep Reinforcement Learning: Applying RL techniques to dynamic optimization problems, particularly in pricing strategies, traffic signal control, and competitive scenarios.

  3. Graph-based Retrieval-Augmented Generation: Advancing RAG systems with dynamic graph structures for improved knowledge retrieval and generation.

  4. Understanding and Analyzing Large Reasoning Models: Probing the reasoning processes of LLMs through interdisciplinary perspectives.

Last updated: November 2026