Research and analysis of hate and other emotions in social media

Abstract

[en] In the course of just a few years, with the massive introduction of social media, people have changed the way they communicate and share experiences dramatically. The global scale that this topic has reached, combined with its rapid expansion, is a historic landmark. However, what do social networks represent in our day-to-day lifestyles? The answer is a double life. Since their launch, a digital pseudo-reality has been created in which thoughts, emotions and privacy can be expressed in detail. This leads us to dump each of society’s concerns into community applications, and if you add the factor of anonymity behind a screen, the result is incendiary. Through this work, it is intended to identify, study and analyze the high level of emotions, mostly negative, that has been flooding social media thanks to the aforementioned anonymity. This process will be carried out by entering the Natural Language Processing field. For this purpose, a study of Hate Speech, Toxicity, Offensiveness and other emotions will be carried out on four datasets, each one with one of these tasks respectively. Using these datasets, three language models, based on Transformers and Deep Learning, will be trained and validated for their future comparison. All of this is performed with the aim of finding the ideal framework for each of the featured tasks, which are based on true-to-life situations. Furthermore, it is intended to find the causes of the inconveniences that the models may present, in a concise and intuitive way for the reader.