Revolutionizing Content Moderation in Digital Advertising: A Scalable LLM Approach

Researchers at Google Ads Safety, Google Research, and the University of Washington have developed a scalable and efficient content moderation system for digital advertising using large language models (LLMs). The system employs heuristic filters and clustering mechanisms to condense the dataset, optimizes resource utilization by reviewing representative ads, and utilizes cross-modal similarity representations for clustering and label propagation. The system has achieved significant reductions in the volume of ads requiring direct LLM review and increased recall compared to traditional approaches. It has the potential to revolutionize content moderation practices across digital platforms.