Vector Based Techniques for Short Answer Grading.

Abstract

Vector-based approaches proved their validity during the past few years as promising techniques for word and sentence representation. Automatic short answer grading is a challenging problem in natural language processing that can reduce a lot of human effort, accordingly research was focused towards exploiting several vector representations to solve this problem. In this paper various sentence representation techniques and wide range of similarity measures are compared and finally a system for short answer grading is presented. The system either outperforms the state of the art systems on different data sets or achieves comparable results.