Combining Data and Human Intelligence through Predictive Visual Analytics to Improve Educational Assessments

Abstract

Assessments are widely used in higher education. It is often the case that most teachers learn the tacit knowledge of creating tests throughout their careers. Despite the proliferation of educational technologies, existing datasets are often still untapped for their potential use to improve test quality. This paper proposes a novel approach to leveraging students' test data to aid teachers in the test construction process. Based on the various student profiles and the domain model learned by the system, teachers can administer their newly created tests through simulation, effectively helping them identify items for revision and improvement without additional effort. The affordances of interactive data visualization are utilized to make such predictive models accessible for the teachers. In addition to the simulation, the system can determine topic coverage, ensuring such test matches a predetermined test blueprint. This system integrates machine intelligence with human intelligence that benefits the teacher in making informed decisions. Students indirectly benefit as they are assessed with better quality tests that could capture their mastery of the domain.