Towards automated quality assurance of patent specifications: A multi-dimensional LLM framework

Towards automated quality assurance of patent specifications: A multi-dimensional LLM framework

Journal article

1 volume.

Despite the widespread adoption of Large Language Models (LLMs) in intellectual property domains, patent evaluation has remained unexplored and continues to rely on labor-intensive manual processes. To address this gap, we propose a multi-dimensional LLM framework for automated patent specification quality assurance. Our framework comprises three specialized evaluation modules: (1) a regulatory compliance module assessing adherence to patent office guidelines, (2) a technical coherence module evaluating logical consistency and technical soundness, and (3) a figure–reference consistency module verifying alignment between textual descriptions and drawings. A final integration step synthesizes module-level defect signals into prioritized, actionable revision recommendations, bridging automated assessment with practical patent drafting and pre-examination workflows. The framework is evaluated on 160 patent specifications, consisting of 80 human-authored CNIPA patents and 80 AI-generated specifications produced by commercial drafting tools. Three evaluation modules achieve balanced accuracies of 99.74%, 82.12%, and 91.2% respectively against expert annotations. Additional analysis examines defect distributions across patent sections, technical domains, and authoring sources, providing further insights into quality patterns of LLM-generated patent specifications.

Crossref

World Patent Information ; 85, June, 2026

World Patent Information

Oxford [England] : Elsevier Ltd., 2026.

English

https://www.sciencedirect.com/science/article/abs/pii/S0172219023000108

External Online Journals > World Patent Information