Automated Assessment System for Subjective Questions Based on LSI

Subjective question is capable of examining the adopting ability of knowledge of the student, but the assessment for it suffers from a number of questions such as trickiness, synonymy and polysemy. This reduces the advantage of subjective question for online exercise. In this paper we explore an approach to automated assessment system for subjective question based on latent semantic indexing. Chinese automatic segmentation techniques and subject ontology are used for transferring the reference answers to a term-document matrix, which is then projected to a k-dimensional LSI space by the statistical technique Singular Value Decomposition to solve the problem of synonymy and polysemy. A reference unit vector is introduced to alleviate the problem of trickiness. The system then concludes the quality of the solution according to the similarity between the projected vectors. The experimental results prove the feasibility of our theoretical architecture and flow for automated assessment of subjective question.