College Course Name Classification at Scale

摘要

Accessing college course content data at scale is often challenging due to a variety of legal and technical reasons. In this study, we classify college courses into course categories using only a college course name as an input. We describe our training data design, training process and report performance and evaluation metrics on two deep learning models- an LSTM and a word sequence-to-sequence models - trained on a three-level hierarchical course taxonomy with a number of course categories ranging from 58 to 2322. Despite scarce input data, the best performing models reach 0.91 accuracy and 88% relevance in quantitative and qualitative evaluations respectively.