The challenge
Scientific knowledge is largely unstructured and static, while existing LLM- and RAG-based approaches do not represent how concepts and expertise evolve over time. This limits systematic analysis, reproducibility, and the identification of knowledge gaps across domains.
The Solution
Knowledge Discovery is a 3-year research initiative led by PI Andrey Ustyuzhanin, focused on building temporal, semantically grounded cognitive maps of scientific knowledge and researcher competences.
The project introduces Temporal Knowledge Hypergraphs (TKH) and Temporal Competence Hypergraphs (TCH) to model the evolution of concepts, methods, and expertise across time, domains, and research communities.
Core innovations include:
- Semantic structuring: Transformation of unstructured data (papers, code, dialogue) into machine-actionable, interpretable graphs.
- Time-aware representations: Explicit modeling of how concepts, methods, and hypotheses evolve over time.
- Competence mapping: Linking scientific outputs (TKH) to researcher profiles (TCH) to support automated expertise search and team formation.
- Task-aware reasoning: Explainable, traceable, user-specific reasoning beyond black-box LLM behavior.
Impact
- Researchers: Faster literature review, discovery of knowledge gaps, and reproducible cross-domain insights.
- Institutions: Competence-aware collaboration mapping and support for strategic hiring decisions.
- AI systems: Benchmarks and foundations for question answering, hypothesis generation, and forecasting grounded in temporal knowledge representations.
Industrial Directions
The solution applies to research-intensive organizations and institutions that manage large volumes of evolving scientific knowledge and expertise. It supports applications such as analytics, expertise discovery, team formresearchation, and AI-assisted scientific reasoning in domains where explainability, temporal context, and reproducibility are critical.
- Artificial Intelligence, Data & Analytics, Decision Support, Digital Platforms & Transformation, Education, Enterprise Software, Innovation, Knowledge & Intelligence Systems, Research & Development, Scientific Computing
