Author Gender Identification for Short Texts

Francisco Antonio Castillo Velásquez, José Luis Godoy Martínez, Jonny Paul Zavala de Paz, José Amilcar Rizzo Sierra, María del Consuelo Patricia Torres Falcón


At present, the possibility of communicating or expressing oneself through an electronic medium is very wide: most users of computers and mobile devices use email, social networks, chats and other tools. One of the problems that has arisen with this form of communication is excess, such as plagiarism, false identity, intimidating notes, and others. The attribution of authorship of texts (AAT) is responsible for answering the question of who is the author of a text, giving some previous examples of that author (training set). A useful process within the AAT is the identification of gender or sex (male, female) and that has been studied by several authors, but mainly for English. The present work proposes a computational model based on lexical characteristics (n-grams) for the identification of the genre for short texts in Spanish. Tests were made with a corpus of text messages on social networks and blogs, obtaining promising results.


Gender identification, machine-learning, n-grams, classification, authorship

Full Text: PDF (Spanish)