As asynchronous online discussions gain a broader usage due to the COVID-19 epidemic, the need for understanding of students' ideas from unstructured textual data becomes more pressing. In this experimental study, we examine the effects of recommendations on message quality and community formation from voluminous online discussions. Drawing on literature from group cognition, knowledge building discourse, and learning analytics, we calculate message quasi-quality index (QQI) scores based on message lexical complexity and topic-related keyword usage by participants in explaining their ideas. Furthermore, we examine the empirical evidence on the relationship between QQI scores and participants' interactions. Finally, we visualize network structures via sociograms and hierarchically cluster participants to identify subgroups. Our analysis of 281 messages finds that recommendations helped participants to write more messages that compared alternative viewpoints and refined preliminary ideas with higher QQI scores. Results also show that recommendations cultivated a sense of collective agency to increase the opportunity for creativity and reduce the likelihood that peripheral participants will be dissatisfied and fail to identify with a community. Theoretical contributions and practical implications are discussed.