Análise da presença de discurso de ódio em vídeos de recomendações do YouTube

Abstract

The analysis of videos aimed at identifying hate speech has motivated the development of many studies in recent years, but mostly with the sole purpose of recognizing them. However, this work aims not only to identify whether this speech occurs in the video or not, but also to assess the level of user exposure to this type of speech while using the YouTube platform. To achieve this, an automatic navigation between recommended videos was simulated, starting from the first one and from one randomly chosen. It was observed that, despite all the filters the platform has for its content, there is still a presence of mainly Offensive content (5%-10% of each video), regardless of the random choice or starting from the first video in the recommended list. Finally, this analysis does not highlight any specific depth of navigation.