TAAL：通过时间自回归对齐缓解生成式推荐中的早期束剪枝

中文摘要 AI 辅助

生成式推荐将物品编码为分层语义标识符（SIDs），并通过自回归解码检索下一个物品。然而，标准的下一个 token 预测未明确覆盖交互序列中存在的多模态转换，导致真实 SID 在早期束搜索分支中易遭受不可逆剪枝。在三个公开基准上，我们发现 91.9%—96.6% 的检索失败发生在前两个解码步骤内。因此，我们提出时间自回归对齐（TAAL）。训练期间，TAAL 从历史转换构建联合（c₁,c₂）软目标，并通过前向 KL 目标对齐早期前缀分布；推理期间，TAAL 用点互信息（PMI）校准候选分数，以降低全局频繁前缀的影响。在 Amazon Beauty、Instruments 和 Yelp 数据集上，TAAL 较标准基线分别提升 NDCG@10 达 39.5%、6.7% 和 28.6%，同时提升全 SID 存活率 3.9%—16.6%。束宽分析进一步显示，相对存活率增益随束宽缩小而增大，在 B=5 时达到 39.4%。

英文摘要

Generative recommendation retrieves items by autoregressively generating semantic identifiers, but beam search may discard a target before its complete identifier is generated. Our preliminary analysis across three benchmarks shows that most missed targets are pruned within the first two decoding steps, highlighting the importance of early prefix retention. However, retaining a target's first-token branch alone is insufficient if its continuation is pruned at the next step: aligning only the first-token distribution improves first-token survival but yields little gain in complete-path retention. This observation motivates joint supervision of early branches and their continuations. We propose Prefix Alignment with Temporal History (PATH), which aggregates transition statistics from the training corpus over recent interactions with exponential decay to construct history-conditioned two-token prefix targets. PATH aligns the model's joint predictions with these targets through forward KL divergence, using a chain-rule decomposition that enables unbiased Monte Carlo estimation. At inference time, PATH reuses the transition statistics for pointwise mutual information (PMI) calibration, reranking completed candidates relative to global prefix frequency. Experiments on Beauty, Instruments, and Yelp demonstrate the effectiveness of PATH, showing consistent improvements in recommendation performance and higher full-SID survival rates.

发表机构

• Harbin Institute of Technology（哈尔滨工业大学）

机构由 AI 辅助整理，请以论文原文为准。