A Concept Map Based Assessment of Free Student Answers in Tutorial Dialogues

Typical standard Semantic Textual Similarity (STS) solutions assess free student answers without considering context. Furthermore, they do not provide an explanation for why student answers are similar, related or unrelated to a benchmark answer. We propose a concept map based approach that incorporates contextual information resulting in a solution that can both better assess and interpret student responses. The approach relies on a novel tuple extraction method to automatically map student responses to concept maps. Using tuples as the unit of learning (learning components) allows us to track students' knowledge at a finer grain level. We can thus better assess student answers beyond the binary decision of correct and incorrect as we can also identify partially-correct student answers. Moreover, our approach can easily detect missing learning components in student answers. We present experiments with data collected from dialogue-based intelligent tutoring systems and discuss the added benefit of the proposed method to adaptive interactive learning systems such as the capability of providing relevant targeted feedback to students which could significantly improve the effectiveness of such intelligent tutoring systems.