Guntuku, Sharath Chandra, Qiu, Lin, Roy, Sujoy Sinha, Lin, Weisi Weisi, Jakhetiya, Vinit, Guntuku, Sharath Chandra, Qiu, Lin, Roy, Sujoy Sinha, Lin, Weisi Weisi, and Jakhetiya, Vinit
In this work, selfies (self-portrait images) of users are used to computationally predict and understand their personality. For users to convey a certain impression with selfie, and for the observers to build a certain impression about the users, many visual cues play a significant role. It is interesting to analyse what these cues are and how they inuence our understanding of personality profiles. Selfies of users (from a popular microblogging site, Sina Weibo) were annotated with mid-level cues (such as presence of duckface, if the user is alone, emotional positivity etc.) relevant to portraits (especially selfies). Low-level visual features were used to train models to detect these mid-level cues, which are then used to predict users' personality (based on Five Factor Model). The mid-level cue detectors are seen to outperform state-of-the-art features for most traits. Using the trained computational models, we then present several insights on how selfies reect their owners' personality and how users' are judged by others based on their selfies. © 2015 ACM.