Virtual mentor and media structuralization theory

Abstract

In the 21st century, e-Learning has been widely used in both academic education and corporate training. However, many e-Learning systems present multimedia instructional material in a static, passive, and unstructured manner, giving learners little control over learning content and process. As a result, higher effectiveness and greater societal potential of e-Learning are hindered. This thesis makes two primary contributions to this trend. From a theoretical perspective, we propose a new concept called "Virtual Mentor (VM)" and a research framework called Media StructuRalization Theory (MSRT). The VM refers to a multimedia-based e-Learning environment that emphasizes interaction, flexibility, and self-direction. The MSRT aims at providing guidance toward effective design and implementation of virtual mentor systems. From a technical perspective, we have developed a prototype VM system called Learning by Asking (LBA), which integrates various information technologies. The major technical innovation is adoption of a novel natural language approach to content-based video indexing and retrieval. We conducted empirical studies to validate a few propositions of the MSRT. The results demonstrated that structuring of multimedia content and the use of instructional videos improved learning outcome significantly. The learning performance of students in an eLearning environment with content structuring and synchronized multimedia instruction is comparable to that of students in traditional classrooms. Our research was enabled by the LBA system, which provides a learner-centered, self-paced, and interactive online learning environment. In order to enhance personalized and just-in-time learning, the LBA system allows learners to ask questions in conversational English and watch appropriate multimedia instructions retrieved by LBA that address learners' interests. Traditional video indexing and retrieval approaches are based on scene changes or other image cues in videos that are not normally available in video lectures. We propose a novel two-phase natural language approach to identifying relevant video clips for content-based video indexing and retrieval. It integrates natural language processing, named entity extraction, frame-based indexing, and information retrieval techniques. The preliminary evaluation reveals that this approach is better than the traditional keyword-based approach in terms of precision and recall.