<p>Malicious entities are conducting illegal activities that undermine fair competition on e-commerce platforms. However, identifying malicious entities remains challenging because malicious entities always camouflage themselves among the sheer volume of benign entities. Even the most advanced automated approaches have to sacrifice precision to obtain an acceptable recall rate due to the camouflages. To discover camouflaged malicious entities efficiently, we propose CamouflageVis, a visual analytics approach that integrates machine insights with human knowledge. Compared with existing methods, our approach incorporates advanced machine learning models to detect malicious entities against their camouflage behaviors. Moreover, we develop a visualization interface to help domain experts obtain deep insights into the camouflages and identify underlying malicious entities. Specifically, we provide a clustering view to show which entities are similar based on the learned embedding features. We then design a feature view and an annotation view to reveal the feature distribution of both benign and malicious entities. We also introduce a community view and an entity view to manifest single-level and group-level aggregation of entities in multi-relational graphs. Finally, we evaluated our approach with two usage scenarios based on real-world datasets.</p>