Primary and Secondary Social Media Source Identification

TL;DR

This paper introduces a deep learning method to trace the origin and distribution history of images across social media platforms, aiding in misinformation verification without relying on metadata.

Contribution

The paper proposes a novel two-stage deep learning approach utilizing cascaded fingerprints to identify social media sources of images, outperforming existing methods.

Findings

Achieves over 84% accuracy in identifying social media chains up to length two.

Does not depend on metadata, making it robust against falsification.

Outperforms existing social media source identification algorithms.

Abstract

Social networks like Facebook and WhatsApp have enabled users to share images with other users around the world. Along with this has come the rapid spread of misinformation. One step towards verifying the authenticity of an image is understanding its origin, including it distribution history through social media. In this paper, we present a method for tracing the posting history of an image across different social networks. To do this, we propose a two-stage deep-learning-based approach, which takes advantage of cascaded fingerprints in images left by social networks during uploading. Our proposed system is not reliant upon metadata or similar easily falsifiable information. Through a series of experiments, we show that we are able to outperform existing social media source identification algorithms. and identify chains of social networks up to length two with over over 84% accuracy.