An Interactive Vehicle Recommender System Based on Decision Trees and Multi-Armed Bandits 2019-01-1079
Recommender systems guide a user to useful objects in a large space of possible options in a personalized way. In this paper, we study how to make recommendations for vehicle purchases. This can effectively reduce the human labor in the traditional setting, where customers get recommended vehicles through conversations with the salesmen in dealerships. Comparing a vehicle recommender system to one in other application domains (movies, music, etc.), we identify two major challenges. First, customers usually only purchase a limited number of vehicles, compared to the number of movies or songs. Thus, it is difficult to obtain rich information about a user's purchase history. Second, the content information obtained about the users (demographic, vehicle preference, etc.) is also very limited during their short stay in the dealership. To address these two challenges, we propose an interactive vehicle recommender system based on the methods of decision tree classification and multi-armed bandit. The decision tree effectively selects important questions for the user and understands the user's preference. With these preference as prior information, the multi-armed bandit algorithm by Thompson sampling efficiently uses the user’s feedback to improve the recommendations. The empirical results show that our system can effectively make interactive vehicle recommendations to the users.
Tong Yu, Ole Mengshoel, Dominique Meroux, Zhen Jiang
Carnegie Mellon University, Ford Motor Company