Scaling Up LLM Reviews for Google Ads Content Moderation
This addresses the scalability challenge of LLM-based content moderation for platforms like Google Ads, though it is incremental as it builds on existing clustering and propagation techniques.
The study tackled the problem of high inference costs and latency when using large language models (LLMs) for content moderation on large datasets like Google Ads by proposing a method that uses heuristics to select representative ads for LLM review and propagates decisions back to clusters, reducing reviews by over 3 orders of magnitude and achieving 2x recall compared to a baseline non-LLM model.
Large language models (LLMs) are powerful tools for content moderation, but their inference costs and latency make them prohibitive for casual use on large datasets, such as the Google Ads repository. This study proposes a method for scaling up LLM reviews for content moderation in Google Ads. First, we use heuristics to select candidates via filtering and duplicate removal, and create clusters of ads for which we select one representative ad per cluster. We then use LLMs to review only the representative ads. Finally, we propagate the LLM decisions for the representative ads back to their clusters. This method reduces the number of reviews by more than 3 orders of magnitude while achieving a 2x recall compared to a baseline non-LLM model. The success of this approach is a strong function of the representations used in clustering and label propagation; we found that cross-modal similarity representations yield better results than uni-modal representations.