Optical image recognition strategy for keyword extraction and page ranking for slide recommendation system

One of the main advantages of the traditional instructor/student interaction is the possibility of receiving immediate instructional feedback within the defined lecture periods. However, the transition into full online, and hybrid pedagogy has gradually replaced instructional feedback with peer feedback. While essential for knowledge transfer among learners, peer feedback largely depends on prior knowledge from learning engagements that may not fully be related to the particular course in question. What we have observed in our learning management system is lots of unnecessarily repeated errors which could easily be averted with the right instruction. To this end, we employed optical image recognition to extract structured query language-based keywords from lecture slides and exercise sheets. Using these keywords, we automatically rank the degree to which a lecture slide suitable matches an exercise task, and thus provide an automatic recommendation to students encountering errors in respective tasks during exercise task engagements. Overall, we achieved a precision of 0.707 for German lecture slides and 0.72 for English lecture slides, which validates our strategy of assisting students with automatic lecture slide recommendations during their online exercise engagements.