Predicting the Finish Before the Draft Ends: Continuous Forecasting of Writing Performance from Process Traces

Abstract

Writing evaluation is essential for improving students’ writing performance and developing their writing skills, yet most automated approaches operate post hoc (i.e., only after the writing task is completed), thereby limiting their ability to deliver timely, in-process feedback that would enable students to take prompt actions to improve their performance. To address this gap, this study investigates the feasibility of predicting final writing quality using students’ in-process writing traces accumulated up to a series of timepoints throughout the course of a writing session. Specifically, we collected writing-process data from 526 K–12 students who completed a 45-min writing task across fi..