To minimize the effort required by human security operators in understanding and resolving attacks on the smart grid cyber-physical system, automated detection, prevention and mitigation tools have been integrated into the infrastructure. However, existing visualization frameworks at command and control centers present information from such tools in a nonintuitive, non-contextual format, reducing the situation awareness and timeliness of decisions. There is a need for frameworks that can contextualize the data in a human-understandable format prior to visualizing. To this end, the paper conducts a high-level review of existing literature, and introduces a conceptual human-on-the-loop framework of three modules: data analyzer comprising Kafka, Apache Spark and R, classifier comprising a deep neural network, and situation-aware decision-maker comprising a learning-based cognitive model. Preliminary proof of concept is shown for data analyzer by applying it to contextualize alerts from multiple photovoltaic systems in Florida.