Social learning has been identified as a key aspect of Massive Open Online Courses (MOOCs), because it holds the promise of scalable peer-based learning and is often the dominant channel in which students and instructors interact. Two key challenges faced by MOOC providers today are the rapid drop-off in participation on these forums over time, and the overflow of information on these forums that makes it difficult for students to navigate through and find desired information. This work performs a large-scale statistical analysis of the decline in participation over time for all courses offered by a MOOC provider over summer 2013, identifying course properties that are correlated with this decline. Further, it develops a model for the process by which the discussions are generated, which is used to design algorithms for ranking newly created forum threads by their relevance.