Towards Attention-aware Concept Map based Review in Video Learning
Dr Ooi Wei Tsang, Associate Professor, School of Computing
COM1 Level 3
MR1, COM1-03-19
Abstract:
Learning requires attention; however, maintaining learners' attention in video learning is challenging. Previous studies have shown that video review has become a spontaneous way to mitigate attention loss and is commonly performed by video learners. However, despite its popularity, how to improve learners' video review experience has yet to be fully explored in previous literature.
This thesis centers on designing, evaluating, and developing techniques to facilitate learners' video review experience. First, we conducted an observational study to understand learners' video review behaviors and challenges in the current video learning environment. We found that several factors impede learners from effective review, and consequently proposed three design goals that need to be achieved to solve these factors: overview support, navigation support, and attention-aware.
With the three design goals, we explored the design of effective video review techniques. Following the established Cognitive Theory of Multimedia Learning (CTML) design principles, we designed two review techniques, CMReview and TSReview. Through the controlled studies, we showed that compared with TSReview, CMReview not only enables better post-review learning performance in understanding long videos but also receives significantly more user preference.
In our second and third work, we focused on facilitating the development of CMReview, which involves two steps: (1) create a hyperlinked concept map from video content and (2) Build the attention-aware component. In the second work, we proposed the ScaffoMapping tool, which helps users create hyperlinked concept maps from video content using automatic concept extraction and timestamp linking features. Our evaluation study shows that users using the ScaffoMapping can create concept maps of significantly higher quality than the baseline system. This work can lower the barrier of creating high-quality concept maps from the video content, which is the primary step of developing CMReview.
In the third work, we implemented the attention-aware component of CMReview by proposing the moment-to-moment attention fluctuation detection technique. The CMReview requires annotating each concept with learners' attention states, which requires fine-grained attention detection because of the various time durations of each concept. To achieve this, we introduced a novel paradigm in psychology, the gradual-onset CPT (gradCPT), for ground truth labeling and trained machine learning algorithms based on EEG signals. Our evaluation study in video learning scenarios shows that our technique outperforms the existing works by achieving sub-second attention state detection with comparable accuracy, which suits the need for building the attention-aware component of CMReview.
Overall, this thesis focuses on the design, evaluation, and development of CMReview that facilitates learners' review of video content. It fills a gap in improving the video review experience, and the proposed system provides a new direction for developing attention-aware systems in video learning.