Exploring facial traits associated with beauty and cuteness based on an alternative forced-choice task

It has been highlighted that female facial attractiveness is divisible into components such as beauty and cuteness. At the same time, in Japanese culture, it has also been suggested that the distinction between beauty and cuteness is ambiguous. This study clarified the similarities and differences between beauty and cuteness evaluations by estimating the utility functions of these for multi-dimensional facial traits using Gaussian process preference learning (GPPL). A total of 53 Japanese female university students provided facial photographs. We embedded each female facial image into the latent representation in the StyleGAN2 network trained on the Flickr-Faces-HQ dataset. Using principal component analysis, the dimension of the latent representations is reduced to an 8-dimensional subspace, which we refer to as the Japanese female face space. The participants were asked to select the most beautiful/cute faces from among the nine images that were synthesized using the pre-trained StyleGAN2 model from within the face space. Based on all preferences, participants' psychological utility functions were estimated using GPPL. Facial traits related to beauty and cuteness were examined based on the averaged utility functions. The results revealed that some facial traits affect both beauty and cuteness. In contrast, some baby schema-related facial traits were associated with cuteness, and some facial features affected only beauty.