Comparison of Supervised Machine Learning Methods for Automated Assessment of Student’s Responses to Dichotomously Scored Items in Financial Literacy Test
The article is devoted to comparing methods of automatic checking of dichotomously scored items in the
financial literacy test. Such approaches to natural language processing as "bag-of-words", n-grams (within
word boundaries), n-grams (across the whole response), and the fastText pre-trained embeddings were
analyzed. The logistic regression was used to classify students’ answers. The analysis was conducted on
the data of ninth-graders from one of the Russian regions. As a result, it was concluded that the "bag
of words" is not suitable for automated evaluation of responses, and it is better to utilize the n-grams
method with vectorization over the whole student’s response.