TDM: A Formal Data Model for Integrating Structured and Unstructured Data

TDM: A Formal Data Model for Integrating Structured and Unstructured Data

Turdakov D.Yu. (ISP RAS, Moscow, Russia)
Mayorov V.D. (ISP RAS, Moscow, Russia)
Ryndin M.A. (ISP RAS, Moscow, Russia)

Abstract

Modern information and analytical systems jointly process documents, information extraction results, knowledge bases, and expert validation. Document formats preserve source content and knowledge graphs capture semantic links, but provenance, verification status, and the lifecycle of extracted facts are usually maintained by external mechanisms. This fragmentation complicates source integration, iterative document processing, and maintenance of domain-specific analytical algorithms. The paper introduces TDM (Talisman Data Model), a formal model for representing documents, extracted facts, domains, and knowledge bases within a single framework. In TDM, documents and knowledge bases are containers of typed facts associated with provenance, verification status, structural position, and domain context. The model uniformly describes source content, extraction results, expert validation, knowledge graph projection, and cross-domain knowledge transfer. The main result is a mathematical model of valid TDM-container states and state-preserving operations. TDM can serve as a formal integration layer for intelligent information and analytical systems; its practical applicability is demonstrated by its use in the Talisman integration platform.

Keywords

Talisman; data model; information-analytical system; document; fact; knowledge base; machine learning; human-in-the-loop; knowledge graph; retrieval-augmented generation (RAG); graph retrieval-augmented generation (GraphRAG).

Edition

Proceedings of the Institute for System Programming, vol. 38, issue 4, part 2, 2026, pp. 193-214

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2026-38(4)-26

For citation

Turdakov D.Yu., Mayorov V.D., Ryndin M.A. TDM: A Formal Data Model for Integrating Structured and Unstructured Data. Proceedings of the Institute for System Programming, vol. 38, issue 4, part 2, 2026, pp. 193-214 DOI: 10.15514/ISPRAS-2026-38(4)-26.

Full text of the paper in pdf (in Russian) Back to the contents of the volume