Violence Detection of Sinhala Image Posts with Autoencoders

Social media has become one of the main sources of information with an increased number of users. Unlike other media, control over the contents in social media is minimal. Due to the less controllability, futile, offensive, and racist contents that trigger hatred and violence can be created and distributed among the community. Image posts are one of the mechanisms in social media to spread an idea, an emotion, or information. Image posts contain both text and visual, hence it has become a powerful mechanism of spreading information. Although some of the existing systems focus on English contents, less research focuses on other languages. This research explains a mechanism to detect Sinhala image posts that trigger hatred and violence using a mechanism in deep learning called autoencoders. The results show that autoencoders can identify offensive and violent content in image posts to a certain extent.