Noise reduction in essay dataset for automated essay grading

Abstract

Marking of a huge number of essays is a very burdensome and tedious task for the teacher and/or trainer. Studies have shown that their efficiency decreases significantly when continuously marking essays over a given period of time. An Automated Essay Grading (AEG) system would be most desirable in such a scenario to reduce the workload of the teacher and/or trainer and to increase the efficiency of the marking process. Almost all the existing AEG systems assume that the relationship between the features of the essay and the essay grade is linear, which may not necessarily be the case. In cases where the relationship between the feature vector and the essay grade is non-linear, none of the existing methods provides a mechanism to capture that and determine an accurate essay grade. This paper proposes a new AEG system, the OzEgrader, that aims to capture both the linear and non-linear relationships between the essay features and its grade, and explains the methodology for noise reduction in the essay dataset.