Título: Natural Language Processing for Identification of Tax-Related Doubts
Autores: João Victor M. de Macedo, Leonardo Andrade, Karla Figueiredo
Resumo: This work aimed to investigate Natural Language Processing (NLP) algorithms to automate the Fale Conosco (Contact Us) channel of SEFAZ-RJ, used to clarify taxpayers’ doubts sent via email. Due to the social distancing situation caused by the COVID-19 pandemic, the channel has become a consolidated means to address tax-related inquiries. Thus, employing Machine Learning/Deep Learning techniques, taxpayers’ doubts were classified with the objective of automating the response process. The results with the BERT-based model achieved an accuracy of 96.6%, contributing to a proposal for reformulating the taxpayers’ inquiry form, as well as indicating more promising techniques to initiate the automation process of the Fale Conosco channel at SEFAZ-RJ.
Palavras-chave: LSTM, BERT, Machine Learning, Deep Learning, Tax Law, Natural Language Processing
Páginas: 6
Código DOI: 10.21528/CBIC2023-116
Artigo em pdf: CBIC_2023_paper116.pdf
Arquivo BibTeX: CBIC_2023_116.bib