A Semi-Supervised Approach for Gender Identification

A Semi-Supervised Approach for Gender Identification Juan Soler-Company1, Leo Wanner1,2 1NLP Group, Department of Information and Communication Technologies, Pompeu Fabra University, 2Catalan Institute for Research and Advanced Studies (ICREA) 1C/ Roc Boronat, 138, 08018 Barcelona, Spain juan.soler@upf.edu, leo.wanner@upf.edu Abstract In most of the research studies on Author Profiling, large quantities of correctly labeled data are used to train the models. However, this does not reflect the reality in forensic scenarios: in practical linguistic forensic investigations, the resources that are available to profile the author of a text are usually scarce. To pay tribute to this fact, we implemented a Semi-Supervised Learning variant of the k nearest neighbors algorithm that uses small sets of labeled data and a larger amount of unlabeled data to classify the authors of texts by gender (man vs woman). We describe the enriched KNN algorithm and show that the use of unlabeled instances improves the accuracy of our gender identification model. We also present a feature set that facilitates the use of a very small number of instances, reaching accuracies higher than 70% with only 113 instances to train the model. It is also shown that the algorithm performs equally well using publicly available data. Keywords: author profiling, gender identification, semi supervised learning, text classification, machine learning 1. Introduction Author profiling and author gender identification are in- creasingly popular research areas in Computational Lin- guistics. The range of potential applications spans from forensic investigations to online marketing studies. In the field of forensic investigations, author profiling can be used, for instance, for pedophile detection in chat rooms, terror- ist activity detection monitoring forum/chat/email data, pla- giarism detection, etc. In the field of online marketing stud- ies, author profiling can be used to analyze customer feed- back data and profile the most active users of the services of online companies. The field of author gender identification presupposes that men and women think, talk and as a result, write differ- ently, such that gender-distinctive linguistic patterns can be extracted to differentiate between genders. In analogy, the broader field of Author Profiling is based on the assump- tion is that authors with similar demographic, social, cul- tural and gender characteristics express themselves follow- ing common patterns that can be analyzed to classify their writings with respect to these characteristics. The majority of works in author profiling use Supervised Learning which requires a sufficiently large corpus of clean, correctly annotated training data. However, in many au- thor profiling tasks, such data is not available. Consider, for instance, forensic applications, where only a limited number of writings of the same author can be counted on, or literature studies, where the amount of written material might be sufficient, but not annotated. In this context, semi- supervised learning (or even unsupervised learning) sug- gests itself as an alternative. The goal of semi-supervised learning is to use unlabeled data and a small sample of la- beled data to learn. In what follows, we present a semi-supervised variant of the k nearest neighbors (kNN) algorithm …

Content truncated to 500 words