Complex event processing (CEP) is a technique for analyzing and correlating large amount of information about events that happen in a timely manner, and being in a position to derive conclusions or even respond to them as quickly as possible. Complex events are raised based on incoming sources productions and according to a set of user-defined rules. However, as the complexity of CEP systems grow, the process for manually defining rules becomes time and resource consuming or even impossible as dynamic changes occur in the domain environment. Moreover, it restricts the use of CEP to merely the detection of straightforward situations than in more advanced fields that require earliness and prediction. Therefore, we present a novel approach for completing the supervision of an unsupervised structure learning task. More precisely, we propose to incorporate an unsupervised technique that derives labels for unlabelled data, depended on their distance. From these results, we automatically generate CEP rules to feed the system. In order to evaluate our approach, we used a real world data-set with data labeled by experts. The evaluation indicates that our approach can effectively complete the missing labels and, in some cases, improve the accuracy of the underlying CEP structure learning system