Inferring Gender of Chinese in Social Networks
Jiaxing Lu, Yang Liu, Bailing Wang · 2017
As an important privacy information1, gender inference task has been cause for concern. Previous studies have focused on feature extraction based on the user's profile and publication and ignored the influence of the privacy information exposed by users when posting messages on gender inference. Thus, we present a new content-based feature, publishing source feature, and design a new weight calculation method for this feature. By experiments, the validity of the feature was proved. In addition, different supervised and semi-supervised approaches including linear support vector machine, Naïve Bayes, Tri-Training and so on. An approach based on linear support vector machine performed best, returning the correct gender for about 85% of the users.