Automatic ranking of essays using structural and semantic features

Abstract

Evaluating an essay automatically has been an area of active research even though there has been a shift to multiple choice answers in many competitive exams. In this paper, we propose an unsupervised technique to rank essays based on the structural and semantic content of the essays. The approach is unsupervised because it makes use of a the complete set of essays to determine the rank of the an individual essay. We purposely avoid deep parsing and the approach is based on use of both structural features of the essay and also the semantic content of the essay. We evaluate the proposed approach on a set of essays submitted to a competition generated from a single prompt. We compare the ranks of the essays with the ranks given by two different human evaluators. The results show a good correlation between the proposed unsupervised algorithm and the human evaluators. The proposed approach, as designed, is independent of any external knowledge base.