Media Verification Bot

Overview

This project combats visual disinformation by automatically detecting manipulated images and videos that spread during elections.

The Challenge

Disinformation campaigns increasingly use manipulated media - from simple edits to sophisticated deepfakes - to spread false narratives. Manual verification cannot keep pace with the volume of content.

The Solution

An automated verification pipeline that:

Monitors Social Platforms: Continuously scans for election-related visual content using keyword and hashtag monitoring.

Analyzes for Manipulation: Applies multiple detection techniques including metadata analysis, compression artifact detection, and deep learning-based detection.

Reverse Image Search: Identifies the original source of images to detect out-of-context usage.

Alerts and Reporting: Notifies fact-checkers of suspicious content with preliminary analysis to speed verification.

Technical Approach

The system combines classical forensic techniques with modern deep learning models. We use TensorFlow for the neural network components and integrate with various platform APIs for monitoring.

Impact

During a recent election period, the bot flagged over 200 manipulated images, 40 of which were subsequently confirmed as disinformation by fact-checking organizations and removed by platforms.