Researchers have developed a novel method for creating cross-domain semantic IDs (SIDs) from user activity on organic feeds to improve ad click-through rate prediction. By leveraging rich behavioral data, their approach significantly boosts AUC scores, with direct feed activity embeddings yielding a +0.213% improvement. They also introduced a quantization technique that reduces storage needs by up to 280x while maintaining performance, and a hierarchical embedding module for end-to-end training. This method proved particularly effective for cold-start users with minimal ad interaction history, showing gains up to +1.522% in industrial ad ranking systems. AI
IMPACT Enhances ad targeting by leveraging user behavior from organic feeds, improving relevance for users with limited ad interaction history.
RANK_REASON The cluster contains a research paper detailing a new method for improving ad ranking. [lever_c_demoted from research: ic=1 ai=0.7]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →