Automatic Handwritten Statics Solution Classification and its Applications in Predicting Student Performance
Hanlung Lin, Thomas F. Stahovich, James Herold · 2020
Abstract Automatically interpreting handwritten solutions to Statics problemsIn previous research, we examined the relationship between students' homework solving processand the correctness of their work. More specifically, we conducted a study in which students inan undergraduate Statics course completed all of their coursework, including homework, quizzes,and exams, using LivescribeTM “Smartpens.” These devices record the handwritten solutions astime-stamped pen strokes, enabling us to see not only the final ink on the page, but also the orderin which it was written. We characterized the solution histories with a number of quantitativefeatures describing the temporal and spatial organization of the work. We found that thesefeatures correlate with problem correctness. However, some of the features require the individualpen strokes to be semantically labeled, for example, as free body diagram strokes or equationstrokes. Manually labeling strokes in this fashion is tedious and time consuming. In this paper,we present techniques we have developed for automatically labeling strokes intro threecategories: free body diagrams, equations, and cross-outs. The later class comprises strokes usedto cross out erroneous work.Our techniques are based on a machine learning approach. Each pen stroke is characterized by aset of 39 features describing its geometric and temporal properties. For example, ink at the top ofthe page and ink written early in the solution process are more likely to comprise a free bodydiagram than an equation. Pen strokes are also characterized with shape recognizers that attemptto identify common symbols, such as equal signs and summation symbols. We evaluated ourtechniques on 810 problem solutions containing nearly 300,000 strokes drawn by 90 students.The approach correctly classified 92% of the pen strokes. This accuracy is sufficiently high thatit should enable accurate automatic evaluation of student work.