Exploratory Analysis of User-Generated Photos and Indicators that Influence Their Appeal
In this paper we analyze whether simple indicators related to photo quality (brightness, sharpness, colour palette) and established content detection techniques (face detection) can predict the success of photos in obtaining more “likes” from other users of photo-sharing social networks. This provides a unique look into the habits of users of such networks. The analysis was performed on 394.000 images downloaded from the social photo-sharing site Instagram, paired with a de-identified dataset of user liking activity, provided by a seller of a social-media mobile app. Two user groups were analyzed: all users in a two month period (N = 122.260) and a highly selective group (N = 3.982) of users that only like <10% of what they view. No correlation was found with any of the indicators using the whole (non-selective) population, likely due to their bias towards earning virtual currency in exchange for ‘liking’. However, in the selective group, small positive correlation was found between ‘like’ ratio and image sharpness (r=0.09, p<0.0001) and small negative correlation between ‘like’ ratio and the number of faces (r=-0.10, p<0.0001).