Structured LLM-based patent comparison across three evaluation dimensions

Structured LLM-based patent comparison across three evaluation dimensions

Journal article

1 volume.

Large language models (LLMs) show promise in document-level comparison but often lack transparency and consistency in judgment. These limitations hinder their use in high-stakes tasks such as patent evaluation, where reliable and explainable comparisons are essential. To address this gap, we propose a structured prompting framework that guides LLMs to compare patents across three evaluative dimensions: Functional Purpose (FP), Technical Uniqueness (TU), and Strategic Value (SV). Prompt refinement improves fairness, stability, and interpretability, but its effects vary across technical domains. Together, these findings position structured comparative judgment as a viable and auditable paradigm for deploying LLMs in high-stakes patent evaluation.

Crossref

World Patent Information ; 84, March, 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