Early prediction of writing quality using keystroke logging

Abstract Feedback is important to improve writing quality; however, provide timely and personalized feedback a time-intensive task. Currently, most literature focuses on providing (human or machine) support product characteristics, especially after draft submitted. However, this does not assist students who struggle during the process. Therefore, in study, we investigate use of keystroke analysis predict quality throughout Keystroke data were analyzed from 126 English as second language learners performing timed academic summarization Writing was measured using participants’ final grade. Based previous literature, 54 features extracted. Correlational analyses conducted identify relationship between quality. Next, machine learning models (regression classification) used grade classify might need at several points The results show that, contrast work, rather limited. None regression outperformed baseline, classification only slightly better than majority class baseline (highest AUC = 0.57). In addition, changed course To conclude, be less clear previously posited.