Zhe Cui (University of Maryland, College Park) and Shivani Rao

NIPS 2017

LLD Workshop 2017


Delivering micro content (small pieces of learning materials) from a course catalog has become important as it helps learners engage quickly with a learning platform on a need-to-know period. In order to recommend videos that can teach skills relevant to a learner, we propose an approach for tagging videos with skills using transcripts. Since, there is no labeled data to train a video-to-skill model directly, we leverage a bunch of techniques that fall under the umbrella of weak supervision. We evaluate our approach using offline metrics like Precision, Recall, F-1 score and online metrics via large-scale A/B testing.