The problem of predicting a user's destination in navigation is being actively studied in the automotive field. In particular, modelizing personalized predictive algorithm from small amounts of a user’s data is a very difficult subject. In this paper, we propose a user-personalized model that can predict a user's destination based on a navigation logs that include the user's destination search, selection, and visiting history. Our proposed method is tested on more than 800,000 data, called IBD (infotainment big data), collected from 43,529 anonymized real customer vehicles, and confirms that it surpasses the conventional prediction accuracy. In addition, our algorithm can works on embedded environments by reducing computational power and memory size compared to deep-learning methodologies. If our method is applied to a real system, even a small AI can predict each user's individual destination well. Furthermore, when the proposed method is used for the location-based service, it is possible to increase the customer satisfaction using our vehicle.

Hyundai Motor Company: Changwoo Chun, Sungsoo Park