Predicting Gender from Human or Non-human Social Media Profile Photos by using Transfer Learning

Abstract

Social media profile photos can demonstrate a variety of information about a person, including her personality, behavior, preference, individuality, and gender. Prediction of gender from social media photos has a number of real life applications such as gender marketing and identification of camouflaged profile photos. Numerous techniques can be applied for determining gender from a user’s profile photos. In this study, we predict a user’s gender from her social media profile photos (i.e., Facebook, Twitter, and Instagram) by using multiple transfer learning models. While conventional methods are straightforward and can only determine gender based on human faces, we propose a novel model that determines gender based on both human faces and non-human pictures (i.e., a flower, animal, cartoon, doll, etc.). The model predicts the gender of a user based on the pattern of sharing profile photos with an outstanding accuracy of 95.75%.