Scholarlink

The creative review process often involves manually sifting through numerous images to identify content that violates advertising policies. The manual creative review team spends a lot of time and resources in detecting harmful content, such as gambling within image creatives. Our objective is to reduce the workload of the team with advanced ML models by classifying the images as gambling or not. For this purpose, we have evaluated CNN-based model (VGG-16), Vision Transformer model (ViT), and LLMs (LLAMA-Vision-11b and LLM2Vec encoder), along with an ensemble of these for the gambling image classification task. We are able to achieve close to 99% accuracy, with misclassification of gambling (FNR) and non-gambling (FPR) being 5% and 1%, respectively. We have also achieved an approximately 95% potential reduction in time, effort, and cost on application of our classification solution.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Large Language Models for Detecting Gambling Advertisement Images to Enhance the Efficiency of the Creative Review Process

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Edward L. Martis,
  • Jayesh Santosh Asawa

摘要

The creative review process often involves manually sifting through numerous images to identify content that violates advertising policies. The manual creative review team spends a lot of time and resources in detecting harmful content, such as gambling within image creatives. Our objective is to reduce the workload of the team with advanced ML models by classifying the images as gambling or not. For this purpose, we have evaluated CNN-based model (VGG-16), Vision Transformer model (ViT), and LLMs (LLAMA-Vision-11b and LLM2Vec encoder), along with an ensemble of these for the gambling image classification task. We are able to achieve close to 99% accuracy, with misclassification of gambling (FNR) and non-gambling (FPR) being 5% and 1%, respectively. We have also achieved an approximately 95% potential reduction in time, effort, and cost on application of our classification solution.