Classificação automática de questões de provas: análise comparativa de algoritmos e aplicação ao Enade

Abstract

Despite the large amount of textual data stored digitally, there are several challenges to process it, which reduce its use. With difficulties attached to processing unstructured data, the studies avoid the exploration and extraction of information from these elements, and many times, they leave out these sets of texts, so the focus turns to other data belonging to the same database. This situation happens with the Brazilian educational data, where the microdata, composed of grades of exams and answers of questionnaires, is frequently evaluated. On the other hand, there is a minor quantity of studies that associates these microdata variables with the content that forms this result, which is the textual data from the exams. Despite the difficulties to extract and process this sort of data, there are some advances in this area. These advances include the automatization of question classification, which makes it practical to analyze these data together. Considering these advances and the lack of textual data exploration provided by the agencies responsible for the Brazilian educational system, this study aims to analyze techniques for processing and classifying textual data in the context of ENADE (Brazilian National Student Performance Exam). After selecting algorithms suitable for this purpose, we have evaluated their performance and accuracy, according to the chosen categorization.