Detecting portuguese and english Twitter users’ gender

• Autor(a) principal: Vicente, Marco Paulo Fernandes
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/11050
• topic: Text mining; Gender classification; Twitter user; Feature selection; Text classification; Mineração de texto; Classificação de género; Utilizador Twitter; Selecção de atributos; Classificação de texto

Existing social networking services provide means for people to communicate and express their feelings in a easy way. Such user generated content contains clues of user’s behaviors and preferences, as well as other metadata information that is now available for scientific research. Twitter, in particular, has become a relevant source for social networking studies, mainly because: it provides a simple way for users to express their feelings, ideas, and opinions; makes the user generated content and associated metadata available to the community; and furthermore provides easy-to-use web interfaces and application programming interfaces (API) to access data. For many studies, the available information about a user is relevant. However, the gender attribute is not provided when creating a Twitter account. The main focus of this study is to infer the users’ gender from other available information. We propose a methodology for gender detection of Twitter users, using unstructured information found on Twitter profile, user generated content, and later using the user’s profile picture. In previous studies, one of the challenges presented was the labor-intensive task of manually labelling datasets. In this study, we propose a method for creating extended labelled datasets in a semi-automatic fashion. With the extended labelled datasets, we associate the users’ textual content with their gender and created gender models, based on the users’ generated content and profile information. We explore supervised and unsupervised classifiers and evaluate the results in both Portuguese and English Twitter user datasets. We obtained an accuracy of 93.2% with English users and an accuracy of 96.9% with Portuguese users. The proposed methodology of our research is language independent, but our focus was given to Portuguese and English users.