Community Detection with Opinion Leaders' Identification for Promoting Collaborative Problem-Based Learning Performance

Abstract

With web-based collaborative problem-based learning, learners could more conveniently cultivate their problem-solving capabilities through autonomous learning. Nevertheless, learners are often guided to solve a target problem by the messages announced by teachers during the collaborative problem-based learning (CPBL) processes. Individual learners often could not effectively absorb such standard messages, thus ignoring the important messages from teachers. This study thus employs the modularity Q function as the fitness function of genetic algorithm (GA) to optimally detect communities and uses PageRank measure to accurately find out community opinion leaders according to the social network interaction data of learners in the CPBL process. Based on quasi-experimental design, this study examines whether learners in the experimental group using the two-step flow of communication through opinion leaders to convey messages for solving a target CPBL mission could more significantly enhance web-based CPBL performance, social network interaction, and group cohesion than learners in the control group using the one-step flow of communication through teachers' messages. Analytical results show learners in the experimental group remarkably outperform those in the control group on learning performance and peer interaction under a CPBL environment. Learners in the experimental group present significantly higher group cohesion than those in the control group.