Automated Short Answer Scoring using Weighted Cosine Coefficient

Scoring short answer question (consisting of at most 100 words) requires a special treatment, particularly on the term weighting process. With such limitation, it is therefore infeasible to have term weighting with term frequency model because the frequency of term appearance is very rare. This paper proposes a new method, named Weighted Cosine Coefficient. The experiment result shows that Weighted Cosine Coefficient method can improve the correlation value on Automated Short Answer Scoring (ASAS) system compared to that of Jaccard Coefficient, Dice Coefficient, Cosine Coefficient, Weighted Jaccard Coefficient, and Weighted Dice Coefficient. The average improvement of the correlation value on the tested data set is 0.56.