An intelligent grading system for descriptive examination papers based on probabilistic latent semantic analysis

Abstract

In this paper, we developed an intelligent grading system, which scores descriptive examination papers automatically, based on Probabilistic Latent Semantic Analysis (PLSA). For grading, we estimated semantic similarity between a student paper and a model paper. PLSA is able to represent complex semantic structures of given contexts, like text passages, and are used for building linguistic semantic knowledge which could be used in estimating contextual semantic similarity. In this paper, we marked the real examination papers and we can acquire about 74% accuracy of a manual grading, 7% higher than that from the Simple Vector Space Model. © Springer-Verlag Berlin Heidelberg 2004.