Transfer learning techniques for cross-domain analysis of posts in massive educational forums

Discussion forums are popular tools in Massive Open Online Courses (MOOCs), used by students to express feelings, exchange ideas and ask for help. Due to the large number of enrolled students, several approaches to automated forum post analysis are emerging for helping instructors moderate and plan their interventions. Such approaches have the common drawback that, when trained on posts from one course or domain, their application on another course or domain is often unsatisfactory. To solve this problem, this chapter introduces a cross-domain text categorization tool that includes transfer learning capabilities for detecting intent, sentiment, confusion and urgency of MOOC forum posts. The tool, based on convolutional and recurrent neural networks, can be trained on a labeled dataset and then adapted to any course or domain by tuning it on a small set of labeled samples. The proposed tool has been experimented and compared with related works.