Beyond Semantic IDs: Finite-Beam Reachability and Continuous-Query Generation in Generative Recommendation

Key points

• This work examines how this output interface constrains whether held-out targets enter finite returned lists, while distinguishing catalog-wide scoreability from empirical top-K support.1
• We introduce two continuous-query paradigms: CQG-Single generates one retrieval query in a single pass, and CQG-AR autoregressively generates continuous residual updates, retaining step-wise decoding while removing the SID item-token vocabulary, code lookup, and discrete beam expansion.1
• These designs thereby eliminate the SID-specific reachability constraint induced by finite-beam decoding and avoid the technical challenge of aligning a discrete SID vocabulary with the pretrained backbone's vector space.1
• These findings motivate decoupling output representation, autoregressive modeling, and retrieval budget when evaluating generative recommendation systems.1

Sources (1)

• [1]Beyond Semantic IDs: Finite-Beam Reachability and Continuous-Query Generation in Generative Recommendation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 04:23 AM

  Generative recommendation (GR) of social media platforms commonly retrieves items by autoregressively decoding discrete Semantic IDs (SIDs) with finite-width beam search.

  This work examines how this output interface constrains whether held-out targets enter finite returned lists, while distinguishing catalog-wide scoreability from empirical top-K support.

Extractive summary: sentences quoted from the sources.