Reasoning rerankers are designed to write a rationale before ranking a candidate slate, and existing systems rely on reading item metadata to do so. That metadata can be verbose and awkward to handle, especially as generative recommenders shift toward more compact item representations.
The new work proposes using Semantic IDs (SIDs) as the sole item representation for such rerankers. SIDs are already used in generative recommenders to represent items compactly, so this approach aligns the reranking stage with the broader recommendation architecture.
By removing the need for separate metadata, the method could reduce complexity and make reasoning rerankers more compatible with modern generative systems. The abstract does not provide experimental details, so the practical gains remain to be seen, but the conceptual shift is notable.