MemeTrace: A Multimodal LLM-Based Agent for Safety-Aware Meme Provenance Analysis

Key points

• Their rapid evolution in response to news events and strong dependence on contextual information make it difficult for platforms and moderators to detect and moderate them effectively.1
• We evaluate MemeTrace on Meme100, a dataset of 100 real-world memes, using both human annotation and automatic metrics.1
• The results demonstrate that MemeTrace substantially outperforms the VLM baseline.1
• More importantly, MemeTrace achieves higher accuracy in detecting unsafe memes, especially those involving disinformation and sensitive content, and remains robust under visual perturbations.1

Sources (1)

• [1]MemeTrace: A Multimodal LLM-Based Agent for Safety-Aware Meme Provenance Analysis
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 10:38 PM

  Online memes are often carriers of unsafe content such as disinformation and propaganda.

  Their rapid evolution in response to news events and strong dependence on contextual information make it difficult for platforms and moderators to detect and moderate them effectively.

Extractive summary: sentences quoted from the sources.