抖音集团旗下巨量引擎首次披露AI广告治理自研大模型

Core Insights - Douyin Group's Juyuan Engine introduced its core strategy of "full-link governance + AI governance by AI" during the 32nd China International Advertising Festival, highlighting its commitment to advertising safety [1] - In Q3 of this year, over 840,000 AI-related advertising violations were intercepted, showcasing the effectiveness of the platform's governance measures [1] - The self-developed multimodal large model can review 90% of materials within 10 minutes, improving review efficiency by 75% compared to traditional methods [2] Group 1 - The Juyuan Engine focuses on "precise identification and strict handling" of advertising content, prohibiting fictional personas and endorsements of products or services [1] - A multi-tiered and diversified handling system has been established for violations, including warning rectifications, restrictions on new openings, account suspensions, and entity removals [1] - The platform has launched a special service guarantee project targeting the elderly and children [2] Group 2 - The self-developed multimodal large model automatically repairs common violations in materials, allowing advertisers to adopt corrections based on their specific situations [2] - For ads that involve violations, the system quickly provides detailed rejection reasons, indicating the specific time or scene in the video where the violation occurs [2]