Impact of an Adaptive Dialog That Uses Natural Language Processing to Detect Students' Ideas and Guide Knowledge Integration

This study leverages natural language processing (NLP) to deepen our understanding of how students integrate their ideas about genetic inheritance while engaging in an adaptive dialog. In Study 1, informed by knowledge integration (KI) pedagogy, we used responses from 1,485 students to test one NLP model to detect the ideas students express when explaining why siblings look similar but not identical and another NLP model to holistically score their response for KI. In Study 2, we used the tested NLP models from Study 1 to design an adaptive dialog that responds to students' detected ideas. We assessed the impact of the dialog on students' level of KI. We embedded the dialog in a web-based unit and implemented it in five middle and high schools with 11 teachers and 610 students. Students' KI scores significantly improved across the unit, and from their initial to revised responses in the dialogs. Consistent with KI, students significantly added differing new accurate ideas. They generally linked their vague ideas to new ideas rather than dropping vague ideas. Two patterns emerged: Students who achieve partial KI form links between new accurate and initial vague ideas; Students who progress to integrated KI distinguish between initial vague and accurate ideas plus new accurate ideas to form varied links. These results clarify that students follow multiple paths to combine their ideas and construct coherent responses while studying a unit featuring adaptive dialogs. They point to designs for adaptive guidance to build on students' ideas and promote integrated understanding.