Automatically extract interpretable topics from online discussion

Abstract

Teachers adopting CSCL often face the challenge of handling massive textual information, and finding it difficult to have a clear grasp of the topics being addressed in the discourse. Topic modeling, an emerging field in machine learning, has the potential to solve this problem by automatically extracting from text collections formal representations of latent topics. However, the interpretation of latent topics is still a challenge, which hinders the use of this state-of-the-art technology from wider use in CSCL contexts. In a recent paper, we put forward a novel topic discovery method, the fLDA model, based on Minsky's Frame theory. This method has the advantage of providing outputs that are potentially more easily interpretable for generating the topic of each thematic cluster. In this paper, we show how fLDA can be used in extracting and visualizing the topics of asynchronous online discourse from four classrooms.