A large language model-based method for trademark similarity analysis in the Brazilian context

Abstract

A trademark aims to uniquely and distinctively identify the products and services offered by a company. It is a key intangible asset, acting as a fundamental tool to prevent unfair competition and strengthen a company’s market positioning. However, the increasing number of trademark applications submitted to the Brazilian National Institute of Industrial Property (INPI) has introduced significant challenges, such as longer processing times, inconsistencies in decisions, and greater complexity in identifying conflicts. In this context, automated methods for trademark similarity analysis become essential to improve the efficiency, reliability, and speed of INPI decision-making processes. This study proposes a method based on Large Language Models (LLMs) to classify and explain the similarity between word marks, following INPI criteria: textual, phonetic, ideological, and market-related aspects. The proposed method is structured into two main components: (1) a classification model to identify conflicts between trademarks, and (2) an explanation model that provides detailed justifications for why two marks are considered similar or not. To develop this method, a dataset comprising real cases extracted from INPI official publications was used. Six open-source LLMs were evaluated on their ability to classify and explain trademark conflicts. The results demonstrated high performance for identifying similarity (accuracy $\approx$99%, F1-score $>$98%, AUC $>$99%). The explanation reports were rated above 4.0 (on a 0–5 scale) by IP specialists. Therefore, our LLM-based proposed method demonstrates potential to modernize the trademark examination process. Ultimately, this study highlights the potential of LLMs to enhance trademark analysis, reduce subjectivity, increase transparency, and make trademark protection more accessible.