Automatic assessment of students' free-text answers with support vector machines

For improving the interaction between students and teachers, it is fundamental for teachers to understand students' learning levels. An intelligent computer system should be able to automatically evaluate students' answers when the teacher asks some questions. We first built the assessment corpus. With the corpus, we applied the following procedures to extract the relevant information: (1) apply the part-of-speech tagging such that the syntactic information is extracted, (2) remove the punctuation and decimal numbers because it plays the noise roles, and (3) for grouping the information, apply the stemming and normalization procedure to sentences, (4) extract other features. In this study, we treated the assessment problem as the classifying problem, i.e., classifying students' scores as two classes such as above/below 6 out of 10. We got an average of 65.28% precision rate. The experiments with SVM show exhilarating results and some improving efforts will be further made in the future.