In internet of vehicles (IoVs), drivers always use navigation systems to plan out routes and optimally navigate real-time road congestion. However, the navigation problem cannot be solved nicely by the present approach due to the emphasis put on one point of view on this problem only. The intelligent transportation systems do not consider the drivers' preferences adequately, and driverless cars do not consider real-time traffic conditions. To solve this dispersion, in this paper, we first configure an image of driver and vehicle, named virtual vehicle, to replace the driver making some decisions in IoV. Then, we propose an incremental learning approach for virtual vehicles based on negative correlation learning algorithm, called divided negative correlation learning algorithm, to obtain the drivers' preference knowledge. In the proposed algorithm, the trained ensemble is divided into three parts, where the first part is trained on a new dataset, the second part is used to retrain the old dataset, and the third part is retained. In the end of the proposed algorithm, we combine the three parts to form a new ensemble. Finally, the experimental results show that virtual vehicles can obtain drivers preference knowledge effectively.