Closing the Knowledge Gap in an Online Learning Community: Network-Analytic Discoveries, Simulation and Prediction

This paper demonstrates that a heterogeneous group of individuals becomes more homogeneous in terms of their speaking abilities in a positive trajectory as they interact more with each other in an online platform. The experimental data comes from a learning platform called ROC Speak (available at rocspeak.com) where individuals can provide written feedback to each other on their speaking skills, in addition to receiving automated feedback. We represent the participants and the interactions among them as a network and use graph signal processing tools to analyze how knowledge transfers across the network. Our results show that the users' performance ratings have an improving trend in addition to getting closer to each other - in other words, the users "pull" each other closer towards the better. We hypothesize that this effect is a result of accumulation of knowledge, as knowledge diffuses across the network through mutually beneficial interactions. These insights motivate modeling and simulation of the user ratings' evolution via consensus-based diffusion dynamics, to synthesize an ideal instance of such knowledge propagation mechanism. Finally, we show that when we couple this "pull" effect with individual performance trajectories, we are able to better predict future outcomes when compared to a baseline predictor that relies on individual trajectories alone - providing evidence that the interaction information holds a key to the participants' learning outcomes at an individual level.