Approximating Human Models During Argumentation-based Dialogues