Automatic Essay Answer Rating Using the Text Similarity Method

Abstract

Assesments are required in order to evaluate the performance of students. A typical method of learning assesments in class is by examination. An essay type exam is a form of assesment where there are no answer choices provided and generally applied to measure the students’ level of understanding of the knowledge. To asses the quality of the essay answers manually is a subjective task as well as time consuming. In this reasearch, we propose an automatic method of assessing essay answers by applying the cosine similarity method.

In this research, the students’ answer document and the correct answer document are used as input. Both documents are then preprocessed and represented in vector form using word2vec. We then measure the similarity between the documents by calculating the cosine similarity of the two vectors. The cosine similarity values are converted back again and used as the final grades. The results of the final grade are then compared to the values given by the instructor to show accuracy of the proposed approach.