A new shared task, ChildSafeAds, has been introduced to identify commercial content within YouTube videos targeted at children. The task utilizes a dataset of 3,360 videos and includes segments from SponsorBlock, transcripts, and linked sales pages. Participants will be tasked with classifying the type of promotion, assigning product categories, and identifying legal risk flags. Notably, the dataset revealed that over 45% of videos failed to use YouTube's in-platform ad disclosure method. GPT-5.4 and GPT-5.6-luna were used to generate initial labels for the dataset. AI
IMPACT This task could lead to better AI tools for detecting undisclosed advertising in child-directed content.
RANK_REASON The item describes a new shared task and dataset for academic research. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
