ABSTRACT
Short answers are brief descriptions consisting of one to three sentences. They are evaluated by comparing the Student Answers with a Reference Answer. But scoring based on the term similarity alone can be inefficient given the fact that each student can answer in their own words. This approach doesn't take account of the contextual meaning of sentences. As a result, the performance of the system will be unsatisfactory. The proposed model focuses on the semantic relation between the student answer and reference answer by using Latent Semantic Indexing. Latent Semantic Indexing helps in representing the semantic similarity between documents in a reduced dimensional space using TF–IDF Document Matrix and Singular Value Decomposition (SVD). TF–IDF matrix reflects term importance instead of term frequency which is an improvement over previous research. The experimental result shows that both the Mean Squared Error and Mean Absolute Error have considerably decreased, compared to the term-based approach.
