AdaQE-CG: Adaptive Query Expansion for Web-Scale Generative AI Model and Data Card Generation

Researchers introduce AdaQE-CG, a framework that automatically generates model and data cards for AI systems with improved accuracy and completeness. The approach combines dynamic query expansion to extract information from papers with cross-card knowledge transfer to fill gaps, accompanied by MetaGAI-Bench, a new benchmark for evaluating documentation quality.

The automation of AI documentation represents a critical infrastructure challenge as the field scales. AdaQE-CG addresses a genuine pain point: model and data cards are essential for transparency and governance, yet manual creation is labor-intensive and inconsistent across platforms like Hugging Face. The framework's innovation lies in its adaptive approach—rather than relying on fixed templates, it iteratively refines queries based on document structure, making it resilient to diverse paper formats and evolving documentation standards.

The broader context reflects growing regulatory pressure and industry recognition that AI transparency isn't optional. The EU AI Act, emerging standards from ISO and IEEE, and user demands for trustworthiness have created institutional demand for comprehensive documentation. Without systematic tools, this becomes a bottleneck as model proliferation outpaces manual documentation capacity.

For developers and platform operators, AdaQE-CG reduces friction in governance workflows. Hugging Face and similar repositories could integrate such tools to automatically improve metadata quality at scale, lowering barriers for researchers and smaller teams to meet documentation standards. The introduction of MetaGAI-Bench also establishes a benchmark for measuring documentation quality, enabling iterative improvement and comparison of approaches.

The public release of code, prompts, and data democratizes access to these techniques. Future developments likely involve tighter integration with model release workflows and application to emerging documentation requirements as AI governance frameworks mature across jurisdictions.

• →AdaQE-CG automates generation of model and data cards by combining context-aware information extraction with knowledge transfer from similar documented systems.
• →MetaGAI-Bench provides the first large-scale expert-annotated benchmark for evaluating AI documentation quality across five dimensions.
• →The framework addresses three core challenges: static templates, incomplete metadata in web-scale repositories, and lack of standardized evaluation protocols.
• →Automated documentation tools reduce governance friction and accelerate compliance with emerging AI transparency standards and regulations.
• →Public code release enables broader adoption and helps standardize documentation practices across AI platforms and repositories.