Classificação de Coleções de NFTs Explorando Metadados e Aprendizagem de Máquina

Abstract

Non-fungible Tokens (NFTs) are digital objects with unique identities and ownership verified via blockchain networks. The digital arts and media industry has embraced NFTs for their secure features, such as defining authorship, transfer, and royalties, which can be programmed into smart contracts. The classification of NFTs is crucial for their commercialization but often relies on the author’s definition, which may be prone to errors, or on expert evaluation. In this work, we analyze NFT collection classes based on metadata extracted from OpenSea, the largest NFT platform. We assess the efficiency of supervised machine learning to identify the most relevant attributes of these collections and classify them into the nine most popular categories on the platform. Our results demonstrate the challenges of automating classification to assist users and platform curators, achieving a promising accuracy of 67% and an F1 score of 72% in the best cases.