Results indicate that SID alignment is conditional rather than universally beneficial, and this note asks whether Semantic ID tokens should be aligned with natural language or specialize into collaborative codes.
Shi-Teng Cao, Zhi-Heng Li· Proceedings of the 20th ACM...· 0 citations
Semantic-ID-based generative recommendation represents each item as a short token sequence and trains an autoregressive model to predict the next item. This note asks whether Semantic ID tokens should be aligned with natural language or specialize into collaborative codes. We report two studies. First, representation p...
Shi-Teng Cao, Zhi-Heng Li· Proceedings of the 20th ACM...· 0 citations
A series of models are explored that connect item Semantic IDs with natural language in a unified representation space and seek to unlock the potential of natural-language chain-of-thought (CoT) reasoning for recommendation, including OneReason.