Comparing atomic feedback with classic feedback on a linear equations task using text mining techniques

In a previous study, we did a crossover experiment with 45 math teachers giving feedback to 60 completed linear equation tasks in two conditions: the semi-automated condition in which they could re-use feedback and were encouraged to write atomic feedback; and a condition in which they wrote classic feedback. Atomic feedback consists of a set of formulation requirements that makes feedback significantly more reusable; instead of writing long pieces describing lots of different mistakes at once, they must (1) identify the independent error occurring and (2) write small, independent feedback items for each error. We already know that the semi-automated system led teachers to give significantly more feedback instead of saving time. This paper now explores the differences and similarities of the provided feedback in both conditions using text mining. We found that the word frequencies and sentiments are similar in both feedback types, while atomic feedback contains fewer abbreviations, more section titles, and more concrete instructions.