Guess your size: A hybrid model for footwear size recommendation

In recent years, online shopping for footwear has rapidly increased. However, the user experience has not been satisfactory because of the size mismatch problem, i.e., customers often fail to choose the right size online. Traditional size selection schemes, including those suggesting that users select footwear sizes according to their past experiences or those based on simple measurements, usually result in a high return rate of up to 35 % . The limitation of the traditional size selection schemes is that they fail to consider (1) the characteristics of foot shapes and (2) the preferences of individual customers. In this paper, we propose a size recommendation framework that is jointly based on 3D (foot and last) features and user preference. First, we report measurement studies of foot shape characteristics based on foot data for 10 K individuals. Our findings reveal that users have diverse foot shapes and different personal preferences regarding size matching. Second, based on our measurement insights, we design a size recommendation model that jointly considers 3D foot models, shoe characteristics and user preferences. We also provide a predictive model that predicts comfort levels for particular parts of the foot based on the given size recommendation. Finally, our data-driven experiments show that the proposed size recommendation improves the size selection accuracy to 92 % , which is a 22 % improvement compared to conventional solutions.