Automatic Topic Labeling for Facilitating Interpretability of Online Learning Materials

Jun Wang, Kanji Uchino

To reduce the cognitive overhead of understanding and organizing online learning materials using topic models, especially for new learners not familiar with related domains, this paper proposes an efficient and effective approach for generating high-quality labels as better interpretation of topics discovered and typically visualized as a list of top terms. Compared with previous methods dependent on complicated post-processing processes or external resources, our phrase-based topic inference method can generate and narrow down label candidates more naturally and efficiently. The proposed approach is demonstrated and examined with real data in our corporate learning platform.