Emprego de técnicas de pré-processamento textual e algoritmos de comparação como suporte à correção de questões dissertativas: experimentos, análises e contribuições

Abstract

This master thesis presents a study of techniques used as support for a correction of essay questions based in an adaptation of string-matching algorithms combined with text preprocessing techniques. The main challenge to design a tool like this is an ambiguity of natural language. To analyze a correction of subjective questions, tests were performed with these algorithms, and a tool have been developed for this purpose. Comparing student responses with response pattern of questions proposed in subjective tests, we analyzed the performance of individual algorithms and a set of pre-processing techniques that are found in the literature, in isolation and combined. Seeking to neutralize specific situations of false negative and false positive, some techniques have been proposed as auxiliary contribution of this work. After analyzing the experiments, the results of similarity index between responses indicate the use of the solution to support the correction of essay questions, and may also be applied in the detection of plagiarism and be integrated to a learning management system.