ABSTRACT
Social media platforms have developed comprehensive community standards to foster safe and inoffensive environments. These community standards cover a range of unwanted content related to crime, violence, hate, pornography, intellectual property, and several other categories. Large platforms employ thousands of content moderators to make judgments on content to take down, but this is not a scalable solution. The constantly increasing volume of user-generated content necessitates the use of automated methods, namely machine learning, to classify content. However, automated classification faces several technical challenges. Some challenges, such as classification accuracy, may be improved over time but other challenges are inherent to the subjective nature of content moderation. At present, automated methods are not good enough to completely replace human judgment.
