Digital violence against women: a phenomenon exploration to understand and counteract from a Data Science perspective

Abstract

Investigations have shown that Violence Against Women is a pervasive problem that has been increasing over the last years. Until a few years ago, it took place both in public and private spaces, but it has now broken into Digital Space adopting more symbolic expressions. Mexican cyberfeminists have fought to put this social problem on public agenda, achieving this past June to legally typify Violence Against Women in Digital Space at the federal level. There have been some important related work from Data Science approaches but mainly on cyberbullying and in the detection of language patterns through supervised algorithms, through social network features, through profile information, a few works on unsupervised learning, and on violence against women. However, it is important to tackle Digital Violence Against Women as a phenomenon with its particularities and separated from cyberbullying. Moreover, it is necessary to study this phenomenon from a gender perspective since all crimes against women are contained by a gender symbolic structure. The hypothesis of this Thesis Project is that Data Science approaches such as Text Mining, Supervised Learning, Time Series Analysis, Natural Language Processing, and Network Analysis can find associations between proposed variables of Spanish-language text data from microblogging social network, Twitter, datasets. The goal of this thesis is to implement Data Science techniques to analyze the Digital Violence Against Women phenomenon in order to achieve the identification of major associations that will let us understand and counteract violent social discourses and structural violence in digital space. The proposed model is composed of several techniques such as Time Series Analysis, Natural Language Processing, and Network Analysis, that are fed by the outcomes of the ensemble between Supervised Classifiers and an Ontological Matcher. Results indicate a higher presence of Digital Violence Against Women for the predicted tweets under the ensemble algorithm in comparison with just the Supervised Learning Algorithms or just the Ontological Matcher. Time Series Analysis shows peaks in Digital Violence Against Women in dates that correspond to days in which the fight for Women’s Rights was positioned. Natural Language Processing confirms the existence of a violent semantic discourse under this phenomenon. And, Network Analysis exhibits generalized individual attacks connected to a structural and systemic problem. Finally, there were four strategies proposed to counteract Digital Violence Against Women, which are based on detection, prevention, and specificity of the phenomenon.