New QRC framework lets AI understand emerging memes in real-time

A team of AI researchers (Shanhong Liu, Rui Cao, Pai Chet Ng, De Wen Soh) has introduced Query Retrieve Conclude (QRC), a zero-shot framework designed to understand evolving multimodal memes by acquiring open-world knowledge in real time. Traditional meme understanding models rely on fixed parametric knowledge that quickly becomes outdated, especially for memes that emerge with new cultural references. QRC solves this by first identifying missing knowledge, then retrieving relevant evidence from the open web, and finally synthesizing that evidence into grounded background knowledge for meme interpretation and detection. The framework was tested on three meme understanding datasets and five meme detection tasks, consistently outperforming zero-shot baselines in knowledge recovery and downstream performance.

To support this work, the researchers also curated a benchmark of recent memes from 2024 to 2026, each annotated with external background knowledge. This benchmark addresses the lack of up-to-date meme datasets that include the necessary context for understanding. QRC's ability to dynamically fetch and integrate web knowledge makes it particularly suited for rapidly evolving meme formats that rely on current events, internet culture, or niche communities. The paper, published on arXiv (2606.05316), represents a practical step toward keeping AI systems culturally aware without requiring constant retraining or manual knowledge updates.