BEAR: Towards Beam-Search-Aware Optimization for Recommendation with Large Language Models

TL;DR

BEAR introduces a beam-search-aware regularization technique for fine-tuning large language models in recommendation tasks, addressing the inconsistency between training objectives and inference behavior to improve recommendation accuracy.

Contribution

The paper proposes BEAR, a novel fine-tuning method that explicitly considers beam search behavior, improving recommendation performance with minimal additional computational cost.

Findings

BEAR significantly outperforms baseline methods on four real-world datasets.

The approach effectively reduces the risk of pruning positive items during beam search.

BEAR incurs negligible computational overhead compared to standard supervised fine-tuning.

Abstract

Recent years have seen a rapid surge in research leveraging Large Language Models (LLMs) for recommendation. These methods typically employ supervised fine-tuning (SFT) to adapt LLMs to recommendation scenarios, and utilize beam search during inference to efficiently retrieve $B$ top-ranked recommended items. However, we identify a critical training-inference inconsistency: while SFT optimizes the overall probability of positive items, it does not guarantee that such items will be retrieved by beam search even if they possess high overall probabilities. Due to the greedy pruning mechanism, beam search can prematurely discard a positive item once its prefix probability is insufficient. To address this inconsistency, we propose BEAR (Beam-SEarch-Aware Regularization), a novel fine-tuning objective that explicitly accounts for beam search behavior during training. Rather than directly simulating beam search for each instance during training, which is computationally prohibitive, BEAR enforces a relaxed necessary condition: each token in a positive item must rank within the top-$B$ candidate tokens at each decoding step. This objective effectively mitigates the risk of incorrect pruning while incurring negligible computational overhead compared to standard SFT. Extensive experiments across four real-world datasets demonstrate that BEAR significantly outperforms strong baselines. Code is available at https://github.com/Tiny-Snow/BEAR-SIGIR-2026 .