Machine learning (ML) and deep learning (DL) provide significant opportunities to enhance our understanding of idea generation in asynchronous online discussions (AODs). Drawing on the interaction analysis model (IAM) as our theoretical framework, we built one baseline ML and three DL systems to automate message classification when assessing collaborative knowledge construction depth in academic AODs. The viability of these systems was demonstrated via four offerings of a traditional online course. We achieved 79% as the highest overall accuracy score across all phases of the IAM. To the best of our knowledge, this study is the first to classify AOD messages across all IAM phases. We contribute to the theory by updating the IAM to better explain how to promote deeper interactions in AODs. Additionally, we provide a methodological blueprint for future research where classifying text is crucial.

In the context of large annotation-based literature discussions, this research examines the affordances of recommender systems on users’ disorientation. Drawing insights from literature on group cognition, knowledge building, and recommender systems, we developed three recommender systems and tested these systems on 136 users. Results indicate that the recommender system with constrained Pearson correlation coefficient similarity metric reduced users’ disorientation and afforded them the opportunity to become better aware of interesting and relevant information based on their needs and preferences without heavy costs in terms of time and effort. With respect to other software conditions, results indicate that users suffered from higher levels of disorientation. These findings counter the claim that annotations reduce disorientation. Theoretical and practical implications are also discussed.