Document Type : Original Article
Author
Associate Professor, Department of Knowledge & Information Science, Allameh Tabata'i University, Tehran, Iran
10.22054/dcm.2026.97362.1338
Abstract
The growing use of large language models (LLMs) in information services has increased demand for outputs that can be processed, linked, validated, and reused by digital content management systems. Yet, machine-readable form does not ensure semantic coherence, factual accuracy, or traceability. This study analyzes challenges in LLM-based structured data generation and examines the contribution and limits of Schema.org. Using a qualitative, descriptive-analytical design and directed documentary content analysis, it organizes the challenges into five dimensions: generation and stability; structure and conformance; semantics and identity; accuracy and evidence; and provenance and traceability. The analysis shows that Schema.org can support shared semantic description through common entity types, properties, and relationships, improving consistency, interoperability, and reuse. It cannot independently validate syntax, guarantee factual accuracy, resolve entity identity, verify relationships, or establish claim provenance. These functions require complementary mechanisms: JSON-LD or RDF for representation; JSON Schema or SHACL for constraint validation; authority data, persistent identifiers, and knowledge graphs for identity and relationship management; and evidence-based retrieval and provenance models for verification and traceability. The study proposes a layered framework in which trustworthiness is assessed through five related dimensions: syntactic validity, structural and semantic conformity, factual accuracy and evidential support, provenance and traceability, and operational reusability. The framework positions Schema.org as a semantic component within a broader digital content management architecture.
Keywords
Subjects