Bots and Gender Detection on Twitter Using Stylistic Features

• Published in: Advances in Computational Collective Intelligence pp. 650 - 660
• Main Authors: Ouni, Sarra, Fkih, Fethi, Omri, Mohamed Nazih
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Author profiling; Bot detection; Gender classification; Machine learning; Style-based features; Twitter
• ISBN: 9783031162091; 3031162099
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-031-16210-7_53

This paper describes our proposed method for the author profiling task at PAN 2019. The aim of this task is to identify the type of a Twitter user (i.e. bot or human). Then, in case of a human, determine its gender (i.e. male or female). Our approach uses a set of language-independent features and it applies machine learning algorithms. After an in-depth experimental study, conducted on English and Spanish datasets, we show that by using a simple set of stylistic information, we can surpass other existing methods that mainly depend on the content of the tweets. For the English dataset, accuracies of 93.06% and 90.04% are obtained for bot an gender classification tasks respectively. Using Spanish tweets, accuracies of 90.53% and 89.11% are achieved for bot and gender detection task respectively.