Classificação automática de documentos utilizando técnicas de inteligência artificial

Abstract

The organization of documents in an institution is extremely important, as it brings practicality and efficiency in retrieval. Manually separating these files according to their type takes many hours of work on the part of the responsible professional. Therefore, this work aims to develop a system capable of classifying institutional documents of the State University of Piauí (UESPI), according to their specific type. The database, initially, was composed of PDF files with the photo of the physical document. For the recognition of the texts contained in the images, the software Tesseract was used with some image processing techniques, in order to improve the performance of text extraction. From this, a database was built with the textual information of the documents and their corresponding class. After defining the database, pre-processing was performed using natural language processing techniques, preparing for the classification phases. The classification phase was carried out by two layers of classifiers, one through regular expressions (seeking to locate the type of document by the title) and another using machine learning models (analyzing the textual content of the text). These layers work with the classification flow going through the regular expressions first, if it cannot identify, it passes to the classifier model. In the final application, an API was built that performs the efficient classification process, using both regular expressions and the machine learning model with an accuracy of 98% in cross-validation.