Testing the effectiveness of automated marking using n-gram method on more algebraic equations in one variable / Pm Dr Arsmah Ibrahim, Nuru'l - 'Izzah Othman and Nor Fuzaina Ismail

An automated marking system for mathematics assessment that is able to execute a line-by-line marking of a student's working as would have been done by a human assessor is much desired since mathematics solutions usually consist of a few lines of working before arriving at the final answer. This paper presents a preliminary work on automated marking of solutions to algebraic equations in one variable using n-gram String Similarity Method. In this method the Dice Coefficient Model is employed to calculate the degree of correctness of the automated marking. The effectiveness of the method is tested via a computer program. The concept of tokens is defined and the Dice Coefficient is evaluated as the ratio between the number of similar n-grams to the sum of n-grams between two expressions that are being compared. The marks given by the average of the dice coefficient is then compared to marks given by manual marking. Manual marking is carried out based on a scoring rubric system. The results indicate that the performance of the n-gram automated marking method is comparable to manual marking when dealing with totally correct solutions. However, the performance in assessing partially or totally incorrect solutions can be improved by improving the efficiency of the program that implements the automated marking and the construction of a more complete answer scheme. This research has managed to show that the n-gram String Similarity Method that employs Dice Coefficient Model has a great potential of providing an automated marking tool to mark mathematics assessments involving equations.