Abstract

Whilst face image in the online dating profiles plays an important role in screening for potential dates, there is also a strong need to protect the privacy of the dating site users. Consequently, the dating websites are in need of effective ways to incorporate user’s face preference into the screening mechanism while reducing the risk of identification from the use of face image. This problem belongs to a broad type of challenges where companies need to balance the two conflicting aspects (face anonymization and face perception) when using face image, such as customer relationship management, hiring decision-making, and perception-based face screening. Building upon the literature on face anonymity from computer science and face perception from social psychology and neuropsychology, the authors propose a Face Anonymity – Perceptibility (FAP) paradigm as a general framework when one attempts to solve three types of tradeoff problems (non-perception, component perception and holistic perception). Four different methods can be applied to achieve particular type(s) of tradeoffs: feature reduction and feature replacement from the computer vision literature, and two face abstraction methods (local abstraction and global abstraction) proposed by the authors and grounded in the social psychology and neuropsychology literature. The authors present an empirical study that tests the FAP paradigm by applying the local abstraction and global abstraction methods to online dating industry, where Type III (holistic perception) tradeoff is desired. The results demonstrate that both methods are effective in balancing the need for face anonymization with the need for maintaining face perception. When evaluated on their predictive performance as online dating screening tool, the proposed Face Abstraction-based Screening (FAS) models can generate an relative improvement of 38.6% (the median across all performance measures) over a model without the abstracted facial information.