Automatic Essay Scoring Based on Coh-Metrix Feature Selection for Chinese English Learners

摘要

Automatic essay scoring can be based on essay's content or form. We believe that both classes of features can reflect some aspects of an essay's quality and they should be combined. In this paper, we use Coh-Metrix and importance measure to extract features that cover a wide range of features relating to the essay's grammatical structure, content, form, cohesion, and so on, and more related to the Chinese English Learners. This is a more complete set of features than those used in the literature and it is expected to better cover an essay's characteristics. SVM and C5.0 classification methods based on these features are used to predict the essay's score. Our experiments show that this set of features can produce good results on Chinese English essays even when we use top 5 and top 15 features with higher importance score.