R^3: Advertisement Compliance Rectification via Group-Relative Experience Extractor and Curriculum Reinforcement

2026-07-08

AI Summary

This work addresses the challenge of automatically correcting policy-violating text in online video advertisements—encompassing both speech transcripts and on-screen text—without compromising the original semantic intent, a limitation of existing approaches. To this end, the authors propose an end-to-end correction framework that jointly optimizes compliance and intent preservation. The framework innovatively integrates a data synthesis mechanism based on a population-relative experience extractor and a curriculum reinforcement learning strategy with hierarchical rewards. It further combines text detection, rewriting, and re-rendering techniques into a unified pipeline. Evaluated on industrial-scale datasets and validated through live A/B tests, the proposed method significantly outperforms current solutions, effectively rectifying violations while better retaining the core advertising intent.

Abstract

Rigorous content moderation is crucial for online advertising but leads to millions of daily rejections. This scale renders manual rectification infeasible, particularly for video advertisements. However, existing safety-driven methods often suffer from aggressive over-editing, which compromises the advertiser's original semantic intent merely to satisfy compliance. In this work, we target the rectification of textual violations in video ads, covering both speech transcripts and on-screen text. We propose R^3, a novel framework designed to harmonize compliance with original semantic intent preservation. Our approach integrates three key innovations: (1) an experience-driven data synthesis framework that bootstraps high-quality supervision via a group-Relative compliance experience extractor; (2) a curriculum Reinforcement learning strategy with hierarchical rewards designed to enforce compliance while maximizing semantic consistency; and (3) a comprehensive video Rectification framework seamlessly integrating text recognition, rewriting, and re-rendering for industrial deployment. Extensive experiments on industrial datasets and online A/B testing demonstrate that R^3 significantly outperforms state-of-the-art baselines, achieving an optimal trade-off between violation rectification and intent preservation.

Problem

• advertisement compliance
• semantic intent preservation
• video ad rectification
• content moderation
• textual violation

Innovation

• Group-Relative Experience Extractor
• Curriculum Reinforcement Learning
• Semantic Intent Preservation
• Advertisement Compliance Rectification
• Video Ad Text Rewriting