Gender Inference for Arabic Language in Social Media

The widespread usage of social media has attracted a new group of researchers seeking information on who, what and, where the users are. Some of the information retrieval researchers are interested in identifying the gender, age group, and the educational level of the users. The objective of this work is to identify the gender in the Arabic posts in the social media. Most of the works related to gender classification has been for English based content in the social media. Work for other languages, such as Arabic, is almost next to none. Typically people express themselves in the social media using colloquial, so this study is geared towards the identification of genders using the Saudi dialect of the Arabic language. To solve the gender identification problem the authors, a novel method called k-Top Vector k-TV, which is based on the k-top words based on the words occurrences and the frequency of the stems, was introduced. Part of this work required compiling a dataset of Saudi dialect words. For this, a well-known widely used social site was relied on. To test the system, we compiled 1200 samples equally split between both genders. The authors trained Support Vector Machine SVM and k-NN classifiers using different number of samples for training and testing. SVM did a better job and achieved an accuracy of 95% for gender classification.