With the enormous scale of massive open online courses (MOOCs), many interesting learning analytics issues are worth studying. Peer grading is one vital issue for addressing the assessment challenge for open-ended assignments or exams while at the same time providing students with an effective learning experience through involvement in the grading process. Most existing MOOC platforms use simple schemes for aggregating peer grades, e.g., taking the median or mean. To enhance these schemes, some recent research attempts have developed machine learning methods under either the cardinal setting (for absolute judgment) or the ordinal setting (for relative judgment). In this thesis, we seek to study both the cardinal and ordinal aspects of peer grading within a common framework. Fi...[ Read more ]

With the enormous scale of massive open online courses (MOOCs), many interesting learning analytics issues are worth studying. Peer grading is one vital issue for addressing the assessment challenge for open-ended assignments or exams while at the same time providing students with an effective learning experience through involvement in the grading process. Most existing MOOC platforms use simple schemes for aggregating peer grades, e.g., taking the median or mean. To enhance these schemes, some recent research attempts have developed machine learning methods under either the cardinal setting (for absolute judgment) or the ordinal setting (for relative judgment). In this thesis, we seek to study both the cardinal and ordinal aspects of peer grading within a common framework. First, we propose novel extensions to some existing probabilistic graphical models for cardinal peer grading. Not only do these extensions give a superior performance in cardinal evaluation, they also outperform conventional ordinal models in ordinal evaluation. Next, we combine cardinal and ordinal models by augmenting ordinal models with cardinal predictions as prior. Such a combination can achieve further performance boosts in both cardinal and ordinal evaluations, suggesting a new research direction for peer grading on MOOCs. Extensive experiments have been conducted using real peer grading data from a course offered by HKUST on the Coursera platform. As another learning analytics issue, dropout prediction, or identifying students at risk of dropping out of a course, is an important problem to study due to the high attrition rate commonly found on many MOOC platforms. The methods proposed recently for dropout prediction apply relatively simple machine learning methods such as support vector machines and logistic regression, which use features that reflect student activities such as watching lecture video and forum activities on a MOOC platform during the study period of a course. Since the features are captured continuously for each student over a period of time, dropout prediction is essentially a time series prediction problem. By regarding dropout prediction as a sequence classification problem, we propose some temporal models for solving it. In particular, based on extensive experiments conducted on two MOOCs offered on Coursera and edX, a recurrent neural network (RNN) model with long short-term memory (LSTM) cells beats the baseline methods as well as our other proposed methods by a large margin.