SIDRec: End-to-End Semantic ID Representation and Modeling for Recommendation
Semantic IDs (SIDs) have emerged as a promising approach for discriminative recommendation by encoding continuous item rep- resentations into compact discrete identifiers. However, existing SID-based methods face two fundamental limitations: semantic mis- match between content semantics and collaborative preferences, and objective mismatch between SID construction and downstream recommendation…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
