Weighted LDA Using Metadata for Extracting Topics Needed by Teachers
Koki Takahashi, Ryosuke Harakawa, Masaki Iisaka, Masahiro Iwahashi · 2020
This paper proposes a method for weighted latent Dirichlet allocation (LDA) using metadata for extracting topics with latent needs by teachers. The proposed method enables us to identify and display contents needed by teachers. Specifically, by improving the original LDA through weighting of word frequencies using metadata that reflects teachers' needs (e.g. Likes), the proposed method enables extraction of topics that are not only common to many contents but also needed by teachers. Experiments using contents on Forestanet, i.e., the largest Web service for teachers in Japan, verify the effectiveness of the proposed method.