A Language-independent gender classifier for Online Social Networks

Designing gender predictor for the Online Social Network (OSN) is receiving considerable attention from the research communities of different domains. However, the gender classifiers proposed by earlier studies for social media content so far, highly rely on the language used by the users for writing the content. It implies that the prediction model trained on one language (say English) will likely fail in identifying the gender of users with other languages (for example, Spanish). The study conducted in this paper aims to identify the features from user content on an OSN, which will assist in devising a Language-Independent Gender Classifier (LIGC). The experiments are performed on the Facebook networking site. The site provides the users with the list of various personal attributes that they may or may not reveal to other users on the network. The presented work collects such information of the Facebook users and carries out rigorous feature analysis to know whether this information varies between men and women on Facebook. Furthermore, several machine learning algorithms including Random Forest, SVM, Naïve Bayes, and kNN have been employed to determine the potential of the proposed feature set. Random forest approach achieves the highest value (70%) of performance metric, AUROC (Area under the Receiver Operating Characteristic). The current study is the first attempt to utilize information revelation for designing a gender identifier for OSN that is independent of the language used by the members.