Merge and Branch: Efficient Tree-Structured Supervised Fine-Tuning of LLMs for Recommendation

arXiv:2610.37973 · cs.IR, cs.AI · Submitted 2026-10-27 · Read on arXiv

cs.IR, cs.AI

Submitted: 2026-10-27

Updated: 2026-11-05

Comments: 13 pages, 8 figures

License: http://creativecommons.org/licenses/by/4.0/

The gist: LLM-based generative recommendation reads long, text-rich interaction histories but generates only short semantic identifiers (SIDs).

Terminology

Abstract

LLM-based generative recommendation reads long, text-rich interaction histories but generates only short semantic identifiers (SIDs). This asymmetry makes supervised fine-tuning inefficient: encoding the history is expensive, yet each history usually provides supervision for only one target. We present Merge and Branch (M&B) to reduce input cost and learn more from each encoded history. Merge compresses each historical item's SID and text into a small number of learned tokens. Branch uses the compressed history to predict multiple target SIDs. To share history computation while keeping these targets independent, M&B organizes each example as a shallow, wide tree. We develop tree FlashAttention to execute this layout efficiently, processing the shared history once and propagating gradients from every branch for end-to-end training. A tree-structured key/value cache extends history sharing to constrained decoding. Experiments on public benchmarks show that M&B outperforms strong recommendation baselines while substantially improving training and decoding efficiency. Online A/B tests on a large-scale e-commerce platform suggest that M&B effectively combines item text with multi-target supervision, yielding statistically significant gains in per-user clicks, orders, and GMV.

Related papers