Large-scale Gender/Age Prediction of Tumblr Users

Abstract--T umblr, as a leading content provider and social media, attracts 371 million monthly visits, 280 million blo gs and 53.3 million daily posts However, it is a challenging task t o target specific demographic groups for ads, since T umblr doe s not require user information like gender and ages during the ir registration. Hence, to promote ad targeting, it is essenti al to predict user's demography using rich content such as posts, images and social connections. In this paper, we propose gra ph based and deep learning models for age and gender prediction s, which take into account user activities and content feature s. For graph based models, we come up with two approaches, network embedding and label propagation, to generate connection fe atures as well as directly infer user's demography. Experimental results on real T umblr daily dataset, with hun dreds of millions of active users and billions of following relati ons, demonstrate that our approaches significantly outperform t he baseline model, by improving the accuracy relatively by 81% for age, and the AUC and accuracy by 5% for gender . Online social media has become a ubiquitous part of our daily life, which allows us to easily share ideas/contents w ith other users, discuss social events/activities, and get con nected with friends. The rich content including text, images, and videos, provide great opportunities for advertisers to champion th eir products to specific groups. In particular, Tumblr offers "n ative advertisement" that allows advertisers to present their sp on-sored posts on the users" interface. Native advertising has gained over 3 billion paid ad impressions in 2015 since it was started in 2012 [1].