의미 커널에 기반한 주관식 문제 채점 보조 시스템
(A) descriptive question marking system based on semantic kernels
- 원문 URL
A conventional supplement subjectivity test marking system only requests mauual marking to examiners through e-mails because of limitations for automatic marking caused by natural language processing and also caused by unstable reliance of marking. This process gave a burden of working load to human examiners and the learner could not check the marking results promtly. In this paper, we utilized the Latent Semantic Kernel methodologies for prompt and impartial marking. We constructed semantic kernels from untagged corpus by using several vector-based models, vectorized examination papers written by learners and examiners, computed similarity of two examination papers of vector form through the sematic kernel, and decided whether or not the paper is correct based on the similarity. We constructed 3,000 questions manually for the experiment, and we extracted about 38,000 indexed terms from about 40,000 newspaper documents to construct sematic kernels. We constructed four semantic kernels for the evaluation. In this paper, we mark the real examination papers by using automatic marking systems based on semantic kernels, and we can acquire about 80% accuracy compared to a manual marking.