Needle in a haystack: Identifying learner posts that require urgent response in MOOC discussion forums

Although massive open online courses or MOOCs have been successful in attracting a large number of learners, they have not been equally successful in retaining the learners to the point of course completion. One critical point of failure in many courses, especially those that use discussion forums as a means of collaborative learning, is the large number of messages exchanged on the forums. The extensive exchange of messages often creates chaos from the instructors' perspective and several questions remain unanswered. Lack of attention and response to urgent messages - those that are critical from the learners' perspective to move forward - becomes a major challenge in this environment. This paper proposes a model to identify "urgent" posts that need immediate attention from instructors. In our analysis, we investigate different feature sets and different data mining techniques, and report the best set of features and classification techniques for addressing the problem of identifying messages that need urgent attention. The results demonstrate the ability to use a limited number of linguistic features with select metadata to build a moderate to substantially reliable classification model that can identify urgent posts in MOOC forums regardless of the course content. The work has potential application across a range of platforms that provide large scale courses and can help instructors efficiently navigate the discussion forums and prioritize the responses so that timely intervention can support learning and may reduce dropout rates.