Reproduce “Proactive Detection and Calibration of Seasonal Advertisements with Multimodal Large Language Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hamid Eghbalzadeh, Shuai Shao, Saurabh Verma, Venugopal Mani, Hongnan Wang, Jigar Madia, Vitali Karpinchyk, Andrey Malevich

Researchers at Meta developed a system for proactively identifying seasonal advertisements using finetuned Multimodal Large Language Models (MM-LLMs) that analyze both text and image content. The system achieved an F1 score of up to 0.98 for single-event detection and 0.97 for multi-event detection, significantly enhancing ad delivery relevance and advertiser satisfaction.

Key insights

  • Multimodal LLMs can effectively learn and detect 'seasonality as a concept' directly from ad content, moving beyond historical performance data or simple keyword matching.
  • Finetuning MM-LLMs on specific classification tasks significantly improves performance over zero-shot approaches, proving essential for accurate seasonality detection.
  • Training models on a 'harder task' (without explicit keywords) enhances their ability to generalize and learn deeper conceptual understanding of seasonality.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.