Abstract
Screening for text-implied structural gaps in policy documents is an important component of public health policy review, particularly when implementation relies on online portals, digital identity verification, remote-care platforms, or self-service processes. However, existing text classifiers, large language model prompting methods, and retrieval-based approaches often provide document-level predictions or general explanations without explicitly representing key elements, relations, and potentially missing links in policy implementation. This study proposes Policy Review Process-Knowledge Graph (PRP-KG), a large language model-enhanced knowledge graph framework for identifying structural-gap review signals across policy clauses, implementation processes, and target populations. PRP-KG segments policy documents into clauses, extracts implementation-related elements using a predefined policy-execution schema, and grounds the extracted entities and relations in source evidence spans. The validated elements are then organized into a three-layer knowledge graph. A graph-consistency feedback mechanism revises missing or schema-inconsistent triples, after which structural-gap patterns identify text-implied gaps in execution specification, accessibility safeguards, and service-support coverage. Experiments on policy-text annotation benchmarks constructed from public sources show that, under the evaluated settings, PRP-KG achieves lower prediction error and more accurate identification of high-priority text-implied review signals than most comparison methods. Controlled analyses indicate comparatively stable performance under the evaluated perturbations and provide evidence-linked outputs that can be inspected by policy reviewers.