A phenotype-driven and evidence-governed framework for knowledge graph enrichment and hypotheses discovery in population data
Adela BÂRA, Oprea Simona-Vasilica
Zenodo (CERN European Organization for Nuclear Research)
Problems Identified (5)
Confirmatory KG Construction: Existing knowledge graph construction methods focus on recovering known relationships rather than discovering novel or context-dependent nodes.
Controlled Hypothesis Discovery: Knowledge graph expansion requires structured hypothesis discovery and controlled inclusion of non-trivial, non-redundant knowledge.
Grounded LLM Claim Generation: LLM outputs in retrieval-augmented settings need better grounding to improve retrieval performance and reduce hallucinations.
Confirmatory KG Construction: Existing knowledge graph construction methods focus on recovering known relationships rather than discovering novel or context-dependent nodes.
Controlled Hypothesis Discovery: Knowledge graph expansion requires structured hypothesis discovery and controlled inclusion of non-trivial, non-redundant knowledge.
Proposed Solutions (5)
Phenotype-Driven Evidence-Governed KG Framework: The paper proposes a phenotype-driven and evidence-governed framework for structured hypothesis discovery and controlled knowledge graph expansion.
GNN-Causal-LLM Unified Pipeline: The approach integrates graph neural networks, causal inference, probabilistic reasoning, and large language models for phenotype discovery, hypothesis generation, and claim extraction.
Multi-Objective Pareto Claim Selection: The framework formulates KG expansion as multi-objective optimization and uses Pareto-optimal selection to choose claims balancing relevance, validation, and novelty.
Phenotype-Driven Evidence-Governed KG Framework: The paper proposes a phenotype-driven and evidence-governed framework for structured hypothesis discovery and controlled knowledge graph expansion.
GNN-Causal-LLM Unified Pipeline: The approach integrates graph neural networks, causal inference, probabilistic reasoning, and large language models for phenotype discovery, hypothesis generation, and claim extraction.
Results (3)
Interpretable Phenotype Discovery:
Context-Dependent Causal Structure Discovery:
High-Quality Evidence-Aligned Claims:
Research Domain
Knowledge graph enrichment and hypothesis discovery in population data