Assessing elementary students' science competency with text analytics

Real-time formative assessment of student learning has become the subject of increasing attention. Students' textual responses to short answer questions offer a rich source of data for formative assessment. However, automatically analyzing textual constructed responses poses significant computational challenges, and the difficulty of generating accurate assessments is exacerbated by the disfluencies that occur prominently in elementary students' writing. With robust text analytics, there is the potential to accurately analyze students' text responses and predict students' future success. In this paper, we present WriteEval, a hybrid text analytics method for analyzing student-composed text written in response to constructed response questions. Based on a model integrating a text similarity technique with a semantic analysis technique, WriteEval performs well on responses written by fourth graders in response to short-text science questions. Further, it was found that WriteEval's assessments correlate with summative analyses of student performance.