A Large Language Model-based Model for Patent Classification

Autores

  • Joao Blanco Federal University of Santa Catarina
  • Karen Norberto
  • Alacides Neto Federal University of Santa Catarina
  • Gustavo Migliorini Migliorini Instituto Euvaldo Lodi de Santa Catarina
  • Alexandre Gonçalves Federal University of Santa Catarina

Palavras-chave:

Patent Classification, Large Language Model, Fine-Tuning

Resumo

 Nowadays, the volume of patents published and made available on a daily basis creates challenges and opportunities in both the academic and private sectors. To this end, one of the key areas in this process is patent analysis. This is concerned with how to provide a theoretical basis centered on a set of tasks that support research and technological development. Among the different tasks, patent classification stands out, mainly due to the large number of classification possibilities. This paper proposes a method for classifying patents by recommending class/subclass rankings using Large Language Models (LLMs). For the evaluation, a patent dataset of public domain was used and different accuracies were measured depending on the number of recommended classes/subclasses (k). The main results indicated a good level of accuracy, reaching values close to 70% for k equal to 6 or 7. Considering the results, the method shows potential for assisting patent examiners in determining the appropriate classifications for a given patent of interest.

Downloads

Não há dados estatísticos.

Publicado

2025-12-05

Como Citar

Blanco, J. ., Norberto, K. ., Neto, A., Migliorini, G. M., & Gonçalves, A. . (2025). A Large Language Model-based Model for Patent Classification. Anais Do Congresso Internacional De Conhecimento E Inovação – Ciki, 1(1). Recuperado de https://proceeding.ciki.ufsc.br/index.php/ciki/article/view/1663