AMI'27: Agentic Mobility Intelligence
OrganizersWeigang Lu (The Hong Kong University of Science and Technology)*, Zheng Liang (The Hong Kong University of Science and Technology), Qinggang Zhang (Jilin University), Changyi Ma (Jilin University), Zhiwen Zhang (Jilin University), Hong Kam Lo (The Hong Kong University of Science and Technology)
*Primary contact: weiganglu314@outlook.com
Abstract
Mobility platforms do not merely predict how people move. Their recommendations influence behavior and change the environment on which future decisions depend. AMI'27 proposes a forum on Agentic Mobility Intelligence, connecting mobility data mining, recommender systems, user modeling, and agentic decision making around this feedback loop. The workshop will examine when agentic capabilities improve mobility decisions, how users and agents respond to recommendations, and how to evaluate the resulting system-level effects. We propose a half-day, in-person workshop at WSDM 2027 for approximately 30 participants. Two invited talks, contributed talks, posters, and a panel will help define a research agenda, identify closed-loop evaluation principles, and build sustained collaboration between WSDM and mobility researchers.
The First Workshop on AutoResearch for Information Retrieval and Recommender Systems
OrganizersYingchen Zhang (University of Chinese Academy of Sciences)*, Mingwei Zhang (Leiden University), Ming Li (Meta), Andy Wang (Meta), Yinglong Xia (Meta), Zhaochun Ren (Leiden University), Ruqing Zhang (Institute of Computing Technology, Chinese Academy of Sciences)
*Primary contact: zhangyingchen23s@ict.ac.cn
Abstract
Research automation in information retrieval (IR) and recommender systems (RS) has traditionally focused on individual stages such as hyperparameter optimization, architecture search, and automated experimentation. Recent advances in large language models (LLMs) and agentic systems are extending this paradigm toward AutoResearch: closed-loop systems that can generate hypotheses, design and execute experiments, analyze results, and iteratively decide what to investigate next. IR and RS provide a particularly compelling testbed for AutoResearch, as research actions are often directly executable, experimentation can be conducted at relatively short time scales, and mature evaluation protocols provide structured feedback. These properties enable automated research systems to not only execute research workflows, but also learn from experimental outcomes and progressively improve their research processes. Despite rapid progress, AutoResearch for IR and RS remains fragmented, with open questions surrounding its scope, methodology, reliability, and evaluation. The First Workshop on AutoResearch for Information Retrieval and Recommender Systems aims to establish a shared foundation for this emerging area by bringing together researchers and practitioners from IR, RS, automated machine learning, AI agents, AI for Science, and industry. The workshop will focus on defining AutoResearch and its boundaries, developing methodologies and infrastructure for automated IR & RS research, establishing evaluation principles for research quality and efficiency, and enabling experience-driven and continually improving research systems. It will also explore how IR & RS can serve as a testbed for broader principles of automated scientific discovery.
Large Language Models (LLMs) for E-Commerce
OrganizersHaoyu Han (Michigan State University)*, Akshat Verma (Airbnb), Bharathi Thangamani (Airbnb), Chun How Tan (Airbnb), Huiji Gao (Airbnb), Zhigang Hua (Meta), Yali Wan (Amazon), Neil Shah (Pinterest), Jiliang Tang (Michigan State University)
*Primary contact: hanhaoy1@msu.edu
Abstract
Large language models (LLMs) are rapidly transforming e-commerce applications, including product search, recommendation, advertising, conversational shopping, customer service, seller tools, and business analytics. E-commerce also provides a challenging research setting due to its large-scale, multimodal, multilingual, and rapidly evolving data, as well as its close connection to user preferences and real-world transactions. This workshop will be the third edition of LLM4ECommerce, following successful workshops at The Web Conference 2025 and KDD 2025. Compared with previous editions, this year we place greater emphasis on emerging agentic and multi-agent e-commerce, including shopping agents, personalized assistants, tool-using agents, and multi-agent interaction and coordination. The workshop will bring together academic researchers and industry practitioners to discuss advances and challenges in methods, applications, evaluation, safety, and deployment of LLM-powered e-commerce systems.
DeepSearch: 1st Workshop on Deep Search and Research Agents for the Web
OrganizersMaolin Wang (City University of Hong Kong)*, Wanyu Wang (City University of Hong Kong), Pengyue Jia (City University of Hong Kong), Qidong Liu (Xi'an Jiaotong University), Chenyi Zhuang (Ant Group), Yu Yang (City University of Hong Kong), Jianliang Xu (Hong Kong Baptist University), Qing Li (The Hong Kong Polytechnic University)
*Primary contact: MorinWang@foxmail.com
Abstract
Search is moving from engines to agents. Systems built on large language models (LLMs) can now plan a research question, issue dozens of searches, read and reason over what they find, and write a grounded answer or a full report, a paradigm widely known as deep search and deep research. These agents promise to change how people find information on the web, but they also raise questions that classic information retrieval (IR) has rarely had to answer. When should an agent search, and when should it stop? How should it be trained when the only reward is a final answer? How should its long trajectories and reports be evaluated? How should search engines serve a client that reads everything and never clicks? And how can the resulting systems be made faithful, safe and affordable at web scale? Building on the community formed by earlier workshops on reinforcement learning and agents for IR, we propose DeepSearch, a new half-day workshop on deep search and research agents at WSDM 2027. Whereas earlier agent-based IR workshops used agents to optimize classic tasks such as ranking and recommendation, DeepSearch treats the search agent itself as a new search paradigm and studies it end to end, from reasoning, planning and learning to deep research, co-design with search engines, evaluation, applications and trust. The half-day, in-person program in Hong Kong will feature a keynote, an industry talk, paper presentations, a live demonstration, a panel and structured group discussions, and will produce a short report on the open problems of this emerging field.
WSDM 2027 Workshop on Contextual Intelligence for AI Agents: Retrieval, Memory, and Reasoning
OrganizersJiayuan Ding (Michigan State University)*, Jiaxing Zhang (Independent Researcher), Huayi Yang (Xi'an Jiaotong University), Xinyi Fan (University of Illinois Urbana-Champaign), Yixin Wang (Stanford University), Qiaoyu Tan (New York University Shanghai), Wenqi Fan (The Hong Kong Polytechnic University), B. Aditya Prakash (Georgia Institute of Technology), Xiangliang Zhang (University of Notre Dame), Subhabrata Mukherjee (Hippocratic AI)
*Primary contact: dingjia5@msu.edu
Abstract
We propose a workshop on contextual intelligence for AI agents, bringing together researchers and practitioners studying retrieval, memory, and reasoning for search, recommendation, and Web interaction. The workshop will examine how agents acquire relevant evidence, preserve and revise knowledge across tasks, and use changing context to guide decisions. Its central goal is to connect these capabilities through shared design questions and evaluation practices, including temporal validity, personalization, reliability, and resource efficiency. The proposed full-day program combines four invited keynote talks, two contributed presentation sessions, and an interactive panel discussion to identify shared challenges and future research directions.
LongSeek: The First Workshop on Long-Horizon Information Seeking
OrganizersChuan Meng (The University of Edinburgh)*, Fengran Mo (Rochester Institute of Technology), Jinghan Zhang (Clemson University), Zhenrui Yue (Google DeepMind), Yan Wang (Yale University), Jimin Huang (The Fin AI), Chen Luo (Amazon), Mohammad Aliannejadi (University of Amsterdam), Michael Bendersky (Databricks), Jeff Dalton (The University of Edinburgh)
*Primary contact: chuan.meng@ed.ac.uk
Abstract
The Workshop on Long-Horizon Information Seeking (LongSeek) is a half-day workshop on systems that search, reason, and interact with users over extended temporal horizons, ranging from single complex queries requiring hundreds of search and reasoning steps to evolving query sequences spanning multiple days, weeks, or months. However, most existing work remains short-horizon; e.g., agentic search systems typically use only a limited number of search calls, and conversational search systems use only a small number of tightly coupled historical turns. Many challenges specific to long-horizon settings remain under-explored, such as context explosion, efficiency bottlenecks, sparse supervision, the lack of evaluation of intermediate steps, and challenges arising in multilingual, multimodal, and domain-specific scenarios, as well as human-centric questions of trust and user control. LongSeek brings together the web search, information retrieval (IR), and natural language processing (NLP) communities around memory and context management, reasoning, optimization, and evaluation for long-horizon information seeking. The program is discussion-oriented, with two keynotes, paper spotlights, a poster session, a panel, and a best paper award.
Management of Agentic Systems: Databases, Graphs and the Future of Web Intelligence
OrganizersChen Zhang (The Hong Kong Polytechnic University), Li Sun (Beijing University of Posts and Telecommunications)*, Yutong Ye (Zhejiang Sci-Tech University), Zhiwei Liu (Recursive), Leong Hou U (University of Macau), Xiaochun Yang (Northeastern University), Ruoming Jin (Kent State University), Charu Aggarwal (IBM), Philip Yu (University of Illinois Chicago)
*Primary contact: lsun@bupt.edu.cn
Abstract
Large language model-based agents are reshaping Web intelligence by changing how users search, integrate, analyze, and act on heterogeneous Web-scale data. Yet current agentic systems still face persistent management challenges: weak grounding, inconsistent long-term memory, brittle tool use, inefficient multi-step execution, limited provenance, opaque reasoning, and insufficient governance. This workshop studies how database research and graph data mining can provide principled foundations for managing reliable agentic systems. Database techniques, including declarative interfaces, query and workflow optimization, transactional state management, provenance tracking, access control, and auditability, can make agent workflows more scalable, consistent, and accountable. Graph techniques complement these foundations by organizing evidence, memory, entities, tools, tasks, and inter-agent interactions into structured representations that support retrieval augmentation, coordination, and interpretable reasoning. This workshop also critically examines the capability boundaries of agentic systems, from information retrieval to scientific research, with an emphasis on safety constraints and responsible governance. The program will feature invited talks, contributed paper and demo presentations, panel discussions, and interactive breakout sessions. By bringing together researchers and practitioners from databases, graph mining, Web intelligence, agentic AI, geometric learning, and trustworthy AI, we aim to establish a shared research agenda, identify benchmark and evaluation directions, and shape the next generation of reliable and well-governed agentic systems for Web intelligence.
GMLLM'27: 3rd Workshop on Frontiers in Graph Machine Learning for the Large Model Era
OrganizersQingyun Sun (Beihang University)*, Ziwei Zhang (Beihang University), Haonan Yuan (Tsinghua University), Xingcheng Fu (Guangxi Normal University), Angelica I. Aviles-Rivero (Tsinghua University), Lingjuan Lyu (Sony AI), Yangqiu Song (The Hong Kong University of Science and Technology), Jianxin Li (Beihang University), Philip S. Yu (University of Illinois at Chicago)
*Primary contact: sunqy@buaa.edu.cn
Abstract
The "3rd Frontiers in Graph Machine Learning for the Large Model Era (GMLLM'27)" workshop focuses on advancing graph machine learning (GML) techniques in the context of increasingly large and powerful models. Graphs offer a principled way to represent structured and relational data, making them essential for capturing complex dependencies in knowledge, systems, and behaviors. As the scale and influence of foundation models grow, graph learning stands at a unique vantage point to enhance model robustness, improve interpretability, and integrate domain-specific relational priors. This workshop explores how graph learning can support emerging needs in knowledge reasoning, temporal and multi-hop inference, and AI systems. It also investigates how advances in representation learning, structure-aware generalization, and efficient graph processing can contribute to trustworthy and scalable AI systems. By convening experts in graph learning, knowledge management, and LLMs, the workshop aims to identify core challenges and opportunities of GML in the large model era.
The First Workshop on Scientific Agents in the Digital World
OrganizersWenrui Zhou (HKUST), Yuhan Liu (MBZUAI), Haolun Wu (Stanford University/Mila), Wenyue Hua (Microsoft AI), Weinan Zhang (Shanghai Jiao Tong University), Qiang Sun (University of Toronto/MBZUAI), Ling Pan (HKUST), Jian Kang (MBZUAI)*
*Primary contact: jian.kang@mbzuai.ac.ae
Abstract
The First Workshop on Scientific Agents in the Digital World brings together researchers developing AI agents that formulate scientific questions, conduct computational investigations, and revise hypotheses in response to evidence. The workshop focuses on how agents connect scientific literature and structured knowledge with executable tools, simulations, and experimental feedback to produce scientifically meaningful and independently verifiable findings. Topics include scientific knowledge retrieval and hypothesis generation, simulation and experimental planning, workflow execution and memory, agent learning and adaptation, human–agent and multi-agent collaboration, and evaluation of scientific validity, reproducibility, and trustworthiness. Closely aligned with WSDM's strengths in information retrieval, knowledge discovery, and data mining, the workshop examines how evolving research questions guide information access and how retrieved evidence informs subsequent scientific decisions. Through invited talks, contributed paper presentations, posters, and a panel discussion, this full-day workshop will connect agent researchers with domain scientists to identify shared challenges, exchange methods and evaluation practices, and develop a research agenda for reliable scientific discovery agents.
