Abstract
We investigated the relationship between individuals' mental health traits and the characteristics of YouTube videos they watch. The mental health traits considered were stress, depression, anxiety, and self-esteem, which were measured using a survey questionnaire. We considered violence shown in a video, brightness and saturation of a video as video characteristics. We utilized the viewing history log data of the participants and analyzed the videos they watched on YouTube using computer vision techniques based on deep learning algorithms. The results revealed that viewers' consumption of violent videos was positively related to stress, depression, and anxiety, but negatively related to self-esteem. Individuals with higher levels of stress, depression, or anxiety tended to view darker videos than those with lower levels of stress, depression, or anxiety.