Automatic Rating of User-Generated Math Solutions

Intelligent tutoring systems adapt to users’ cognitive factors, but typically not to affective or conative factors. Crowd-sourcing may be a way to create materials that engage a wide range of users along these differences. We build on our earlier work in crowd-sourcing worked example solutions and offer a data mining method for automatically rating the crowd-sourced examples to determine which are worthy of presenting to students. We find that with 64 examples available, the trained model on average exceeded the agreement of human experts. This suggests the possibility for unvetted worked solutions to be automatically rated and classified for use in a learning context.