Authors:
Jonathan Nau
1
;
Aluizio Haendchen Filho
1
;
Fernando Concatto
1
;
Hercules Antonio do Prado
2
;
Edilson Ferneda
2
and
Rudimar Luis Scaranto Dazzi
1
Affiliations:
1
Laboratory of Applied Intelligence, University of the Itajai Valley (UNIVALI), Rua Uruguay, 458, Itajai, Brazil
;
2
MGTI, Catholic University of Brasilia (UCB), QS 07 - Lote 01, EPCT, Bl. K, sala 248, Taguatinga, Brazil
Keyword(s):
Automatic Essays Scoring, Argumentation Mining, ENEM, Machine Learning.
Abstract:
This paper presents an approach for grading essays based on the presence of one or more theses, arguments, and intervention proposals. The research was developed by means of the following steps: (i) corpus delimitation and annotation; (ii) features selection; (iii) extraction of the training corpus, and (iv) class balancing, training and testing. Our study shows that features related to argumentation mining can improve the automatic essay scoring performance compared to the set of usual features. The main contribution of this paper is to demonstrate that argument marking procedures to improve score prediction in essays classification can produce better results. Moreover, it remained clear that essays classification does not depends on the number of features but rather on the ability of creating meaningful features for a given domain.