Grading Multiple Choice Questions Based On Similarity Measure (Published)
Recent findings show the poor performances of students in Multiple Choice Questions (MCQ) examinations. This could be attributed to the administration of questions on application of knowledge rather than on fundamental of knowledge. Candidates doing these examinations at times choose options that are close to the actual answer but get zero (0) as the reward. The objective of developing a system that rewards a candidate based on the approximation of an option chosen not on the exactness of the option to the answer was therefore formulated in this study. 500 MCQs with their model answers, students’ answers and scores obtained from two universities in Nigeria were collected. Jaro similarity measure was used to compute the degree of similarity between the model answers and the student answers. Results of the experiment show an average deviation of 13.3 marks. The adoption of this method of grading would be very beneficial in evaluating learners on e-learning platforms. The result is encouraging but could be improved upon with semantic similarity measure hybridized with string similarity.
Keywords: E-learning, Examinations, Fuzzy Logic, Jaro, MCQ, approximation, average deviation, exactness, semantic, similarity measure