Aplicação da inteligência artificial para conhecimento e registro digital de produtos com ênfase na classificação tributária

Artificial Intelligence is increasingly present in our lives, from devices that are part of our daily use, such as our smartphones, computers, and virtual assistants, to services we use, such as online shopping, digital social media interactions, portfolios and investment analysis, weather forecasting, risk evaluation, and many others. AI is being applied to activities that are laborious, costly, and pose risks to human health, such as activities in hostile environments or those that humans cannot perform due to time, capacity, or cost constraints, such as predictive disease analysis, behavioral analysis, audits, and others. One of the activities that is laborious and costly, due to the thousands of Brazilian tax laws and regulations, is taxation-related activities. According to IBPT (2020), since 1988, 829 tax regulations are edited per business day. In this work, we researched the use of AI in this area, the existing works and applications, and proposed new technical products that could facilitate tax-related activities for taxpayers. We found research gaps and the lack of tools aimed at taxpayers, where existing tools seek to assist the government in curbing tax evasion. Given this cenario, a bibliographic product, three technical products related to specific objectives, and finally, a technical product related to the main objective of this work were created. A literature review article was created, which was published in the USP International Conference in Accounting 2022; XML Invoice Reading and Processing Algorithm; Database of Products, People, and Tax Data; Cadastral and Fiscal Data Processing and Export Algorithm; and, for the main objective, an ICMS Tax Classification Algorithm based on Artificial Intelligence. These products were implemented and unified into a single program. In the measurements performed, the technical products presented good results, such as the reading, processing, and data persistence time in the database, with a rate of 165 documents per second, the precision of the Artificial Intelligence Algorithm, which reached 97.2% accuracy, obtained after 2.42 minutes of processing and training on a database of 2,839,599 tax movements.