PATENTWRITER: A Benchmarking Study for Patent Drafting with LLMs

TL;DR

This paper introduces PATENTWRITER, a comprehensive benchmarking framework for evaluating large language models in patent abstract generation, demonstrating their high-quality output and potential to improve patent drafting processes.

Contribution

It presents the first unified benchmark for LLMs in patent writing, including diverse evaluation metrics and analysis methods, advancing research in automated patent document generation.

Findings

LLMs can generate high-quality patent abstracts surpassing domain-specific baselines.

The benchmark assesses robustness, style, and downstream task performance of LLMs.

Modern LLMs like GPT-4 show strong potential in patent drafting applications.

Abstract

Large language models (LLMs) have emerged as transformative approaches in several important fields. This paper aims for a paradigm shift for patent writing by leveraging LLMs to overcome the tedious patent-filing process. In this work, we present PATENTWRITER, the first unified benchmarking framework for evaluating LLMs in patent abstract generation. Given the first claim of a patent, we evaluate six leading LLMs -- including GPT-4 and LLaMA-3 -- under a consistent setup spanning zero-shot, few-shot, and chain-of-thought prompting strategies to generate the abstract of the patent. Our benchmark PATENTWRITER goes beyond surface-level evaluation: we systematically assess the output quality using a comprehensive suite of metrics -- standard NLP measures (e.g., BLEU, ROUGE, BERTScore), robustness under three types of input perturbations, and applicability in two downstream patent classification and retrieval tasks. We also conduct stylistic analysis to assess length, readability, and tone. Experimental results show that modern LLMs can generate high-fidelity and stylistically appropriate patent abstracts, often surpassing domain-specific baselines. Our code and dataset are open-sourced to support reproducibility and future research.