Abstract
In this study, we tackle industry challenges in video content classificationby exploring and optimizing GPT-based models for zero-shot classificationacross seven critical categories of video quality. We contribute a novelapproach to improving GPT's performance through prompt optimization and policyrefinement, demonstrating that simplifying complex policies significantlyreduces false negatives. Additionally, we introduce a newdecomposition-aggregation-based prompt engineering technique, which outperformstraditional single-prompt methods. These experiments, conducted on realindustry problems, show that thoughtful prompt design can substantially enhanceGPT's performance without additional finetuning, offering an effective andscalable solution for improving video classification.