Graph-Alignment Approach towards Identifying Gaps in Student Answer

Abstract

Formative assessment aims to provide hints regarding quality of student answers. The hints are in the form of gaps or irrelevant entities in the student answers on comparison with the corresponding model answers. The main objective of the present work is to extract gaps in short answers provided by students. In this work, each of the student and model answers for a given question are transformed into graphs by extracting different relations present in the answers. The nodes of the resultant answer graphs are aligned by considering the similarity of neighborhood topologies of a pair of nodes in addition to the node-node similarity. This leads to the effective extraction of gaps in the form of meaningful phrases in student answers, rather than mere words. Evaluation metrics for reporting performance of the proposed task have been formally defined in this paper. We have compared our proposed methodology with a word-word alignment based baseline system. The proposed methodology outperforms the baseline system in a significant margin.