Recommend related discussion forum posts to students in the small private online course

Computer Organization course from Beihang University adopted hybrid teaching mode based on Open edX platform - project related materials were published online through videos and e-texts, an external grader was integrated online to feedback the evaluation of students' submissions, a virtual online lab system was built to support the remote hardware experiments, and a discussion forum was used to enhancing the communication between participants. The benefits of active online discussion have been noted in literature on evidence-based education in general.

In this poster, the authors report last three iterations' work about how to support continual engagement in the online discussion forum. The survey launched at the end of Fall 2016 semester had shown that, most students said the discussion forum could help in their learning. But due to its asynchronous nature, students could not receive the reply in time, which affected their enthusiasm greatly. Furthermore, the highly effective habit is searching first. If they've searched the forum and could not find a question or topic that is the same or similar to theirs, it's time to post their own.

Here we set up a recommender system that could feed back to student with similar posts when they want to post in the forum. The research includes how to build a corpus related to our course, how to represent the post in a vector and then to compute the distance between the posts. First, we use the Chinese wiki, the "Principles of Computer Organization" text book, the online course materials and forum posts in 2015 semester to build the course related corpus. And then, the posts are represented as vectors using the word2vec technology. Finally, the Word Mover Distance (WMD) and keywords Jaccard Distance between the posts in the forum are calculated. The poster will display some preliminary results: we calculated the course forum posts' WMD and professional contexts keywords Jaccard Distance between Fall 2015 and 2016; when input one forum post in the Fall 2016, the posts in Fall 2015 are set as similar ones if their keywords Jaccard Distance < 0.6, then recommend the top5 most similar posts that are sorted by WMD; and the precision of recommended posts could reach around 0.25 which was evaluated by the course designer.