One of the most challenging tasks in the field of Educational Data Mining (EDM) is to cluster students directly based on system-student sequential moment-to-moment interactive trajectories. The objective of this study is to build a general temporal clustering framework that captures the distinct characteristics of students' sequential behaviors patterns, that tracks whether a student's learning experience is "unprofitable," and can identify such an individual as early as possible so personalized learning can be offered. The central idea of our framework is based on Dynamic Time Warping (DTW), which calculates distance between any two temporal sequences even with different lengths. In this paper, we explore both the original DTW and our proposed normalized DTW to generate distance matrix and apply Hierarchical Clustering to the resulted distance matrix. To fully evaluate the power of our temporal sequential clustering framework, we calculate distance matrix at three types of granularity in the increasing order of: problem, level, and session across three training datasets. As expected, results show that clustering moment-to-moment temporal sequences at problem granularity is more effective than level and session granularity. In addition, our proposed normalized DTW is more effective than both original DTW and the baseline Euclidean distance. [For the full proceedings, see ED596512.]