Automated Methods for Projecting Class Difficulty Onto Lecture Material

Automated Methods for Projecting Class Difficulty Onto Lecture Material

In the quest to turn learning analytics into meaningful stories, we must look at the scalability of the underlying methodology. This presentation explores the ability of natural language processing to automatically determine the difficulty of recorded lectures based on student assessment materials. Skip-grams and Long-Short-Term Neural Networks are used to convert class text into a semantic word embedding and correlate assessment and lecture material. This workflow allows teachers to see student performance directly projected onto their lecture material. Initial results show that utilizing a skip-grams projection significantly reduces over-fitting and qualitative reviews of key lecture passages offer evidence of moderate to high model accuracy.

Speaker:
Tanner M Phillips, Indiana University, PhD Student