This study addresses the challenge of fostering interpersonal relationships and trust in large-scale online courses (MOOCs), where traditional methods often fall short. It introduces group-level indicators to evaluate social activity and participation structures, using exponential random graph modeling to analyze networks of interaction among learners. The study finds that non-facilitated groups exhibit direct reciprocity, while highly facilitated communities show more triadic-level exchanges and less super-posting activity. These insights reveal how different facilitation strategies impact social activity and community dynamics in MOOCs.