Using Text Mining to Extract Information from Students' Lab Assignments

We use text mining as a part of formative assessment in one generation of the first year undergraduate students enrolled in our Statistics course. Practical lab exercises were done in R and RStudio. During the semester, students wrote weekly assignments. There were 11 assignments. Assignments were prepared in Rmarkdown, and students added necessary code and text to the Rmarkdown document, knitted the document and submitted the generated html files. We wanted to identify students whose submissions were not substantially different from the assignment (substandard work) and groups of students who submitted similar texts (possible collusion or plagiarism). Additionally, we wanted to compare results of text mining with students' final grades. Text analyses were done in R using packages XML, RCurl, RWekaand tm. A document term matrix was created with trigrams. For each assignment, distance between submitted texts was calculated as the Euclidian distance. Hierarchical cluster analysis with complete linkage was used to identify similarities between students' assignments. The initial assignment was included for comparison. Clusters of assignments with very small inter assignment distances pointed to similarities of assignments. Students whose work was most distant from others were found to be either excellent, or failing students.