Agrupamento no OpenRe ne: analise de desempenho para agrupamento de itens com variações ortográficas em documentos fiscais

Abstract

The Electronic Invoice (NFe) has modernized the countrys tax administration, reduced costs and bureaucratic obstacles, and improved control by tax administration agencies. However, despite data input veri cation, NFes may contain incorrect information, ma king it di cult to identify fraud or tax evasion. With the increasing volume of data to be veri ed, the need to use arti cial intelligence and data mining techniques to de tect fraud becomes evident. The aim of this study was to compare di erent clustering methods provided by the OpenRe ne platform to group similar entities with orthographic variations in a dataset of retail NFes. Considering the peculiarities of the data and the speci c project requirements, the obtained results revealed that the k-nn method, em ploying the Levenshtein distance function with a radius of 4, along with the key collision method using the Metaphonic3 and Cologne phonetic functions, demonstrated satisfac tory performance, indicating that they are promising options. For optimal results, it is recommended to con gure the k-nn method using the PPM with a radius of 1, thus achi eving excellent precision and sensitivity. Correct data analysis can generate signi cant gains in terms of e ciency and cost reduction in various areas of activity, such as inven tory management and logistics. However, the study presented some limitations, such as the sample size, which could be expanded in future studies to ensure more robust and generalizable results, and the possibility of exploring other machine learning techniques using other platforms to improve the models performance.