Ad Meta Generation for Accurate Conversion Probability Prediction

The patent introduces a generative AI model as an intermediary component that processes unstructured advertiser submission information and advertisement creative images into structured meta information. This mediator enables the system to maintain operational simplicity while achieving high prediction accuracy by automatically extracting and transforming relevant features from unstructured data sources.

The patent replaces traditional manual feature extraction and structured data processing mechanisms with an AI-based automated system. The generative AI model substitutes for complex manual processing pipelines, enabling the system to handle unstructured information efficiently and achieve accurate conversion probability predictions without proportionally increasing system complexity.

The patent implements batch prompting where the generative AI model processes multiple advertisement items simultaneously in advance, preparing meta information for all advertisements before the actual prediction process. This preliminary batch processing significantly reduces the time required for individual predictions while maintaining high processing quality through the AI model's comprehensive analysis capabilities.

The patent uses vectorization to create condensed representations (vectors) of advertisement information that capture essential features. By working with these compressed vector representations rather than processing full advertisement data individually, the system achieves fast and efficient predictions without sacrificing the quality of analysis, thereby improving productivity while reducing processing time.