The goal of proactive contact tracing is to diminish the spread of an epidemic by means of contact tracing mobile apps and big data analysis. Finding superspreaders as has been used in Japan and Australia during the early days of the COVID-19 pandemic has proven effective as backward contact tracing can pick up infections that might otherwise be missed. In this paper, we formulate a proactive contact tracing problem to identify the superspreaders using maximum-likelihood estimation, graph traversal and deep learning algorithms. This problem is challenging due to its sheer combinatorial complexity, problem scale and the fact that the underlying infection network topology is rarely known. We propose a deep learning-based framework using Graph Neural Networks to iteratively refine the supervised learning of proactive contact tracing networks using smaller infection networks and to identify the superspreader. By optimizing the graph traversal and topological features for deep learning, proactive contact tracing strategies can be developed to contain superspreading in an epidemic outbreak.