Shared agenda
Establish agentic data mining as a shared research agenda across communities.
01 — Overview
From passive understanding to autonomous data mining.
As large language models evolve from passive interfaces into autonomous systems capable of planning, tool use, and iterative execution, structured data mining is shifting from task-specific prediction toward end-to-end analytical workflows.
AgentDM brings together researchers in agentic AI, data mining and knowledge discovery, data management, graph and temporal learning, information retrieval, and trustworthy AI to study how agents can ground user goals, plan analyses, invoke tools, discover and predict patterns, verify intermediate results, and communicate reliable evidence from heterogeneous structured data.
How can LLM agents reliably perform end-to-end structured data mining?
- 01Ground
- 02Plan
- 03Invoke tools
- 04Discover & predict
- 05Verify
- 06Explain
Reliable workflows
Identify architectures, representations, reasoning and planning mechanisms for reliable analytical workflows.
Beyond final answers
Develop evaluation principles assessing grounding, planning, tool execution, verification, robustness, and provenance.
Lasting community
Build a cross-community network producing reusable resources, benchmarks, and future shared tasks.
02 — Scope
Six themes of interest
We welcome methods, systems, benchmarks, empirical studies, resources, reproducibility studies, and position papers that place LLM agents as substantive components of end-to-end data-mining workflows.
Foundations & Architectures
Principles, architectures, and representations for autonomous agents over tables, databases, graphs, time series, spatiotemporal data, knowledge graphs, interaction networks, and structured Web data.
Grounding, Reasoning & Planning
Grounding goals in schemas; task decomposition, semantic parsing, model/tool selection, workflow synthesis, reflection, and iterative execution over heterogeneous schemas and evolving APIs.
Discovery, Prediction & Decisions
Agent-driven pattern and knowledge discovery, anomaly detection, relational reasoning, forecasting, recommendation, trend discovery, automated experimentation, and hypothesis generation.
Tools, Memory & Collaboration
Databases, retrieval and Web search, SQL/Python environments, Web APIs and analytical models; memory, state management, human–agent interaction, and multi-agent analytics.
Trust, Verification & Responsibility
Verifying intermediate computations and conclusions: statistical validity, uncertainty, provenance, reproducibility, interpretability, robustness, privacy, fairness, and governance.
Evaluation, Benchmarks & Deployment
Datasets, benchmarks, simulators and protocols for end-to-end agentic mining: generalization, scalability, cost, reliability, distribution shift, and Web-scale deployment.
+ Cross-cutting scientific questions
- How can an agent transfer analytical strategies across tables, graphs, databases, and temporal data without reducing every task to unconstrained code generation?
- How should long-horizon workflows represent uncertainty and propagate errors when an early grounding or tool-selection mistake can invalidate later conclusions?
- What intermediate state — plans, programs, evidence, diagnostics, provenance, self-checks — makes autonomous analysis auditable and reproducible?
- How should agents adapt when schemas, distributions, APIs, or tools change during deployment?
- When should an agent act autonomously, and when should it defer to a human analyst?
03 — Call for Papers
Submit your work
Full submission guidelines, page limits, and the submission portal will be released soon. Final policies follow the WWW 2027 workshop and Companion Proceedings rules.
Contribution types
- Full research papers
- Short / emerging-idea papers
- Benchmark & resource papers
- Reproducibility & evaluation studies
- Position papers
- Non-archival track*
* Subject to WWW policy, a clearly separated non-archival option may accommodate relevant work already published or under review elsewhere.
Review process
- Double-blind review for archival submissions
- ~3 reviews per paper with explicit conflict-of-interest handling
- Criteria: technical quality, relevance, rigor, reproducibility, clarity of evidence, and potential to stimulate discussion
- Contribution-appropriate criteria for benchmark, reproducibility, and position papers
- All accepted papers are presented as spotlights + posters
04 — Timeline
Important dates
All deadlines are 23:59 Anywhere on Earth (AoE, UTC−12). Dates marked tentative will be confirmed once the WWW 2027 workshop schedule is final.
05 — Program
A workshop built for interaction
Full-day, in-person. Substantial time is devoted to participant interaction rather than a mini-conference format.
Spotlights & PostersEvery accepted paper gets the stage, then deeper poster discussion.
Research Speed-DatingTimed pairings across complementary expertise.
Breakouts & PanelGroup outputs stress-tested by panelists and audience.
Benchmark RoadmapOpen problems and shared tasks, released publicly.
Tentative schedule · Local time (Dublin)
Talk Papers Interactive Panel Break
06 — Keynotes
Invited speakers
Perspectives spanning data mining, language agents, structured reasoning, and large-scale industrial systems. Announcements coming soon.
07 — Team
Organizers
An international team across academia and industry, spanning data mining, LLM agents, recommendation, structured & temporal data, multimodal learning, and trustworthy AI.
Program Committee
Program Committee — to be announced.
08 — Beyond the day
Community outcomes
- The AgentDM Research Roadmap of open problems
- An Evaluation Checklist for agentic structured data mining
- A curated public index of datasets, environments & tools
- A shortlist of candidate shared tasks for a future benchmark