Meta (Facebook)

This project aims to address the challenge of detecting inflammatory, misinformed, and propagandist multimodal content on social media. The research focuses on three key objectives: (i) identifying the techniques used in such manipulative posts, (ii) creating a comprehensive set of annotation protocols and guidelines for labeling both text and images in these posts, and (iii) developing a machine learning algorithm capable of detecting these harmful posts.

Visual elements, such as sensational images and Internet memes, have been identified as powerful tools in spreading misinformation and propaganda. By constructing a corpus of manipulative images and text, and by building a detection algorithm, the project seeks to equip the research community with the tools needed to combat misinformation. All intellectual property generated by this research will be open-sourced, encouraging further innovation and collaboration.

Additionally, the project aims to enable proactive measures against misinformation by allowing the identification and demotion of content with high deception scores before it can go viral. This could facilitate more efficient fact-checking and hate speech detection, providing clear explanations for content moderation decisions. Through this work, the project aspires to set new standards for understanding and mitigating the impact of deceptive social media content.