Note-taking behavior is often regarded as one of the important external representations of information use activities in the completion of search tasks, especially learning-related search tasks. In this study, we conducted a user experiment to investigate whether and when users' note-taking behavior occurs through the search activities in each query segment. The results indicate that combining users' dwell time on content pages and Activity Transition (AT) features resulted in prediction accuracy of 86.9%. Furthermore, this study also found that a Hidden Markov Models (HMMs) was able to identify note-taking behavior in real time with good performance, especially in the first query segments in each search session, with the accuracy of 76.65%. These findings provide the effective approaches for real-time monitoring of users' note-taking behavior based on search activities, and contribute to the design of search systems that better support note-taking behavior and task completion during search.