A virtual keyboard control apparatus and control method are disclosed. The virtual keyboard control method includes collecting typo data including a desired input key and an erroneously input key, analyzing the typo data, and updating the key mapping of at least one virtual key of the desired input key and the erroneously input key by using the analyzed result. According to the present disclosure, it is possible to implement the optimum key mapping in which a typo is less generated for each user through Reinforcement Learning of Deep Learning by using the reward called a reduction in a typo rate.