The keys to writing

Subtitle:a writing analytics approach to studying writing processes using keystroke logging

Written text plays an important role in our daily communication, work, and learning. During their studies, students are trained to write these texts in a clear and effective way. However, providing personalized feedback on writing can be challenging and time-consuming for teachers. Therefore, the current thesis proposes and develops an automated tool based on keystroke logging. With keystroke logging all keys that a student presses on a keyboard during a writing task, are logged. By automatically interpreting these keystrokes and presenting the results in an understandable manner, students can gain automatic feedback on their writing, and teachers can gain insight in the strengths and weaknesses of a student's writing. This was done in several steps. First, we identified what students and teachers need to know about students’ writing processes. This resulted in a wide variety of indicators, ranging from the number of words written, to the amount of critical thinking and the spread of revisions over time. Thereafter, we identified to what extent this information could be extracted from keystroke data. Here, it was shown that keystroke data differs too much between student and task to automatically predict writing quality. However, we managed to automatically detail students’ efforts made in revising the document. Automated machine learning models enabled us, for instance, to distinguish lower-level typo (surface) revisions from higher-level (deep) revisions. Lastly, these automated models were used to visualize students’ revision processes in a so called learning dashboard. These visualizations were constructed together with writing teachers. The evaluations showed that the dashboard could be used to gain meaningful insights. Specifically, the dashboard could be used to make students more aware of their own revision processes as well as those of others. The dashboard displays multiple approaches to writing, which could be used by students to improve their writing, for example by making higher-level revisions or by choosing a more effective revision strategy. To conclude, by combining input from the users with automated models, we showed how keystroke data could be transformed into understandable insights with educational applications.

ISBN:978-94-6416-083-3

Publication year:2020

Keywords:Doctoral thesis

Accessibility:Open