Hateful Content Identification and Censorship on Facebook

Abstract

AI (Artificial Intelligence) must be able to grasp information holistically, like humans do, to improve its effectiveness as a tool for identifying hate speech. When we look at a meme, for example, we do not separate the words from the visuals; we get the meaning. Machines, on the other hand, have a huge difficulty since they are unable to evaluate each word and image separately. They must be able to combine these many modalities and understand how the meaning shifts when they are provided together. The goal is to create early in the categorization process tools that integrate the numerous modalities present in a piece of material. The team created a Google Chrome Extension that scrapes posts from a user's Facebook account and utilizes Machine Learning Models to determine whether the messages are hostile. The Google Chrome Extension blurs posts containing hostile material if the content is hateful. The text sentiment analysis model classifies the combined content from OCR and picture captioning model as hateful and non-hateful, and the google extension blurs the hateful post on the user's Facebook account.