The Dynamics of Retraction in Epistemic Networks