In education, exams are used to asses students’ acquired knowledge; however, the manual assessment of exams consumes a lot of teachers’ time and effort. In addition, educational institutions recently leaned toward distance education and e-learning due the Coronavirus pandemic. Thus, they needed to conduct exams electronically, which requires an automated assessment system. Although it is easy to develop an automated assessment system for objective questions. However, subjective questions require answers comprised of free text and are harder to automatically assess since grading them needs to semantically compare the students’ answers with the correct ones. In this paper, we present an automatic short answer grading method by measuring the semantic similarity between the students answer and the correct answer. A semantic network was built to represent the relationship between the words of the two texts to calculate semantic similarity. Representing a text as a semantic network is the best knowledge representation that comes closest to human comprehension of the text, where the semantic network reflects the sentence's semantic, syntactical, and structural knowledge. Several features were extracted from the semantic network and used as input to train the support vector machine (SVM) model to predict the degree of the targeted semantic similarity. The proposed method was tested on the Mohler dataset that is publicly available online. The obtained results were evaluated and reported in terms of Pearson correlation and root mean squared error (RMSE) where it achieved 0.63 and 0.83 respectively. The proposed method outperformed all previous methods on the used dataset.
