Model of conceptual change for INQPRO: A Bayesian Network approach

Constructing a computational model of conceptual change for a computer-based scientific inquiry learning environment is difficult due to two challenges: (i) externalizing the variables of conceptual change and its related variables is difficult. In addition, defining the causal dependencies among the variables is also not trivial. Such difficulty stemmed mainly because conceptual change is an implicit mental model restructuring process, which occurs as a result of confrontation with new knowledge; (ii) to model the process of conceptual change across time by merely observing student interactions could be misleading. This is largely because within the computer-based learning environment, students are granted the freedom to explore and evaluate their ideas. To ease these challenges, this study began with proposing variables of conceptual change and subsequently employing a Bayesian Network model to capture the causal dependencies between the proposed variables as well as the evolving patterns of conceptual change. To obtain the optimal model, two conceptual change Bayesian Network models were proposed and integrated into INQPRO before they can be empirically evaluated via a field study. In this study, interaction logs from 96 were collected and preprocessed before feeding into the proposed models. The accuracy of each Bayesian Network model was measured by matching the results of posttest and interview with the predicted conceptual change outcomes of each model. The empirical findings supported the notion that the conceptual change Bayesian Network model with self-regulation node is a better solution to predict a student's conceptual change in INQPRO.

► We created two models of conceptual change using Bayesian Networks for a scientific inquiry learning environment.
► The models were evaluated using dataset and field studies.
► The model with self-regulation node performed better as compare to that without the node.