End-to-End Semantic ID Generation for Generative Advertisement Recommendation
Researchers have developed UniSID, a novel framework for generating Semantic IDs (SIDs) in generative recommendation systems, specifically for advertisement recommendation. This new approach addresses limitations in existing methods like Residual Quantization by jointly optimizing embeddings and SIDs in an end-to-end manner, preventing semantic degradation and error accumulation. UniSID also incorporates a multi-granularity contrastive learning strategy and a summary-based ad reconstruction mechanism to capture finer-grained semantics. Experiments show UniSID improves Hit Rate metrics by up to 4.62% in downstream advertising scenarios compared to current state-of-the-art methods. AI
IMPACT Enhances generative recommendation systems by improving semantic ID generation, potentially leading to more effective advertisement targeting and user engagement.