THE USE OF TEXTS MINING IN THE SUPPORT TO CORRECTIONS OF DISCURSIVE QUESTIONS IN A HIGHER EDUCATION INSTITUTION

The present research has as main objective the computational development with the use of techniques of Texts Mining for the task of correcting the dissertative questions online, making it possible to provide the diminution of the subjectivity in the evaluation of the discursive questions of the students. The set of data used for the experiments is based on 15 discursive computational questions belonging to the basic course cycle of the Engineering area. The proposed methodology is supported by three major phases: 1) Application of pre-processing techniques and representation of documents according to the Bag of words approach, with term-frequency weighting scheme; 2) Carrying out the processing of texts by comparing the terms contained in the answers with those of the template by means of measures based on terms and editing; 3) Confrontation of the numerical results obtained with the evaluator's correction notes, investigating the hypothesis that the means of the real and estimated scores are equal by means of the T-Test, as well as analysis of the percentage absolute mean error (MAPE, in Portuguese) between such subsets. The results obtained indicated a high adherence to the hypothesis that the averages of the actual vs. estimated data are the same, especially for the tokens-based measurements. The accuracy was of the order of 84.2% for Cosine in the bigram model. Thus, the main result of this work is the design of a TM model to support the evaluation process of discursive issues in the distance learning environment.