Logo image
Back
Component-Aware Case Retrieval for Professional Ethics with Ontology-Constrained LLM Extraction
Book chapter

Component-Aware Case Retrieval for Professional Ethics with Ontology-Constrained LLM Extraction

Christopher B. Rauch and Rosina O. Weber
Case-Based Reasoning Research and Development, pp 203-219
2027
DOI: https://doi.org/10.1007/978-3-032-33865-5_13
Featured in Collection :   Drexel's Newest Publications

Abstract

case-based reasoning component-aware retrieval LLM knowledge acquisition ontology-constrained extraction professional ethics
Professional ethics boards determine whether conduct conforms to established codes by comparing fact patterns against prior opinions and ethical code provisions, but identifying relevant precedents in a growing archive remains predominantly a manual process. Even when opinions are available in electronic form, text search is unlikely to surface complex ethical relationships between cases, and similarity scores based on embeddings provide no explanation for which aspects of two cases are related. We describe ProEthica, an applied CBR system that uses ontology-constrained LLM extraction to transform NSPE Board of Ethical Review cases into structured representations with nine components drawn from the computational ethics literature. The extraction produces named ontology entities for each component, embedded independently for retrieval, so similarity scores are traceable to the specific entities that produced them. Users can examine shared Principles, analogous Roles, differing Actions, or any combination of the nine components across retrieved precedents. Ground truth validation across 119 cases shows that structured methods recover cited precedents at 3 to 8 times the random baseline, outperforming section-based embedding.

Metrics

1 Record Views

Details

Logo image