Clusterização de precedentes de IRPJ no CARF

• Autor(a) principal: Costa, Fabiano de Castro Liberato
• Data de Publicação: 2023
• Outros Autores: Martinez, Antonio Lopo, Klann, Roberto Carlos
• Tipo de documento: Artigo
• Idioma: por; eng
• Título da fonte: Revista de contabilidade e organizações
• Texto Completo: https://www.revistas.usp.br/rco/article/view/197181
• topic: Corporate income taxation; Clustering; Taxation jurisprudence; Administrative Council of Tax Appeals; Tributação de rendimentos corporativos; Clusterização; Jurisprudência tributária; Conselho Administrativo de Recursos Fiscais

The objective of this study was to cluster judgments of the Administrative Council of Tax Appeals (CARF) related to corporate income tax (IRPJ) rendered between 2016 and 2020, employing machine learning (ML) techniques for the clustering of textual documents. The analysis resulted in 13 unique clusters, an unprecedented finding in the tax accounting literature in Brazil. This identification is relevant for the CARF, taxpayers, tax administration, and accounting and tax professionals involved in accounting and tax issues related to the IRPJ. The ML algorithms used proved efficient in solving complex natural language processing (NLP) problems, such as creating vector representations of terms and identifying themes in unstructured data, providing valuable contributions to understanding controversial IRPJ issues in light of administrative case law. The clustering of precedents translates into greater accessibility and analysis of patterns in judgments, facilitating decision-making in tax accounting.