Call for WSDM Day Presentations

Agentic AI and LLM-based systems are increasingly being used to support hypothesis generation and reasoning over complex models. Integrating web search and data mining allows these agents to continuously draw upon open scientific resources, digital libraries, and web knowledge graphs, thereby grounding hypotheses in the latest accessible evidence. Such integration accelerates workflows in fundamental sciences as well as user facility operations wherein researchers must make data-driven decisions regarding experimental and simulation design and control. By embedding reasoning into agents with web-mined knowledge, autonomous systems can help scientists frame and test ideas quickly and efficiently.

Topics in this area include:

• LLMs and agentic models for hypothesis generation and knowledge extraction that leverage large-scale web-mined corpora and open-access repositories.

• Query formulation and expansion using scientific knowledge bases and web-scale search engines.

• Multi-agent systems for problem solving in scientific workflows enriched by open data and community-contributed resources.

• Foundation models for single and multiple scientific domains, trained and evaluated on mined web and literature data.

• Privacy, trust, and interpretability in autonomous reasoning systems that rely on heterogeneous, web-sourced knowledge.

Laboratories increasingly integrate robotics and AI for high-throughput experimentation and autonomous sample handling. Embodied intelligence can enable closed-loop operation of particle accelerator beamlines, synthesis robots, and advanced microscopes, reducing human intervention and increasing reproducibility. Web search and data mining provide complementary resources by enabling embodied systems to draw on prior experimental data, shared lab protocols, and open-access knowledge to guide adaptive control. There is also strong interest in developing digital twins of experimental systems that integrate web-accessible datasets to guide exploration of large parameter spaces.

Topics in this area include:

• Robotics and mechatronics for operating experiments, informed by mined online design repositories and shared experimental logs.

• Closed-loop control in complex laboratory settings that leverages streaming data and searchable metadata collections.

• Benchmarking and validation through open science datasets, test collections, and evaluation methodologies made discoverable via data mining.

• Digital twins and online data streaming for linking simulation to experiment, enriched by knowledge integration from web and community datasets.

Search and planning algorithms play a central role in scientific discovery pipelines, from designing materials with desired properties to planning experiments. When integrated with web search and data mining, such algorithms can exploit open repositories of molecular structures, experimental data, or prior results to guide decision-making. In laboratories, adaptive planning algorithms can maximize information gain from limited resources, such as beam time and sample availability, while leveraging web-accessible experiment registries. This area also includes the operational task of optimal planning and scheduling of experiments at scientific facilities, aided by searchable databases of facility usage and constraints.

Topics in this area include:

• Tree search, graph algorithms, and combinatorial methods that integrate mined data from scientific publications and web repositories.

• Filtering and re-ranking results based on relevance and criticality using web-scale knowledge graphs and citation networks.

• Reinforcement learning and adaptive experimental design augmented with external data mined from past studies.

• Planning and scheduling under uncertainty in constrained environments, supported by searchable facility schedules and open metadata.

Science experiments and simulations often generate petabytes of multi-modal data. Example data sources include particle collisions, climate simulations, or high-resolution microscopy images and spectra. In-situ and edge computing strategies must be used to perform reduction and compression at the source. Web search and data mining play an essential role in discovering, indexing, and linking these massive datasets with open-access resources, enabling cross-facility comparisons and meta-analysis. These capabilities are crucial for processing data from user facilities and making large-scale simulation models usable in practice.

Topics in this area include:

• In-situ and edge computing for real-time reduction of experimental data, linked to searchable web repositories for downstream use.

• Streaming analytics for large-scale experimental datasets integrated with data mining for anomaly detection and trend discovery.

• Compression, summarization, and representation learning for scientific data, with searchable embeddings accessible via open web interfaces.