Cross-domain Author Gender Classification in Brazilian Portuguese

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1227–1234 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1227 Cross-domain Author Gender Classification in Brazilian Portuguese Rafael Felipe Sandroni Dias, Ivandr´e Paraboni School of Arts, Sciences and Humanities (EACH) University of S˜ao Paulo (USP) {rafaelsandroni,ivandre}@usp.br Abstract Author profiling models predict demographic characteristics of a target author based on the text that they have written. Systems of this kind will often follow a single-domain approach, in which the model is trained from a corpus of labelled texts in a given domain, and it is subsequently validated against a test corpus built from precisely the same domain. Although single-domain settings are arguably ideal, this strategy gives rise to the question of how to proceed when no suitable training corpus (i.e., a corpus that matches the test domain) is available. To shed light on this issue, this paper discusses a cross-domain gender classification task based on four domains (Facebook, crowd sourced opinions, Blogs and E-gov requests) in the Brazilian Portuguese language. A number of simple gender classification models using word- and psycholinguistics-based features alike are introduced, and their results are compared in two kinds of cross-domain settings: first, by making use of a single text source as training data for each task, and subsequently by combining multiple sources. Results confirm previous findings related to the effects of corpus size and domain similarity in English, and pave the way for further studies in the field. Keywords: author profiling, gender classification, cross-domain profiling 1. Introduction Author profiling is the computational task of predicting de- mographic characteristics of a target author based on the text that they have written. For instance, by analysing text from social networks or customer’s product reviews, we may infer an author’s gender, age, personality traits or other kinds of information. Author profiling tasks have been a popular NLP research topic, and have regularly featured in the PAN-CLEF shared task series (Rangel et al., 2016; Rangel et al., 2017; Rangel et al., 2018). Practical applica- tions include marketing research, on-line fraud detection, copyright and plagiarism investigations, among others. Author profiling models will often follow a single-domain approach, that is, a model is trained from a corpus of la- belled texts in a given domain, and it is subsequently vali- dated against a test corpus built from precisely the same do- main. Although single-domain settings are arguably ideal, this strategy gives rise to the question of how to proceed when no suitable training corpus (i.e., a corpus that matches the test domain) is available, and in which case we may have to resort to so-called cross-domain author profiling. Cross-domain author profiling has become a popular research topic in recent years (Rangel et al., 2016; Medvedeva et al., 2017), and our own work focuses on the issue of cross-domain gender classification in the Brazil- ian Portuguese language. The focus on gender is moti- vated by the observation that gender-labelled corpora are widely available, and that …

Content truncated to 500 words