LIME：基于链接的用户-物品交互建模与解耦 XOR 注意力，实现高效测试时扩展

View PDF HTML (experimental)

Abstract:Scaling large recommendation systems requires advancing three major frontiers: processing longer user histories, expanding candidate sets, and increasing model capacity. While promising, transformers' computational cost scales quadratically with the user sequence length and linearly with the number of candidates. This trade-off makes it prohibitively expensive to expand candidate sets or increase sequence length at inference, despite the significant performance improvements. We introduce \textbf{LIME}, a novel architecture that resolves this trade-off. Through two key innovations, LIME fundamentally reduces computational complexity. First, low-rank ``link embeddings" enable pre-computation of attention weights by decoupling user and candidate interactions, making the inference cost nearly independent of candidate set size. Second, a linear attention mechanism, \textbf{LIME-XOR}, reduces the complexity with respect to user sequence length from quadratic ($O(N^2)$) to linear ($O(N)$). Experiments on public and industrial datasets show LIME achieves near-parity with state-of-the-art transformers but with a 10$\times$ inference speedup on large candidate sets or long sequence lengths. When tested on a major recommendation platform, LIME improved user engagement while maintaining minimal inference costs with respect to candidate set size and user history length, establishing a new paradigm for efficient and expressive recommendation systems.

Comments:	NeurIPS 2026
Subjects:	Information Retrieval (cs.IR); Machine Learning (cs.LG)
Cite as:	arXiv:2510.18239 [cs.IR]
	(or arXiv:2510.18239v4 [cs.IR] for this version)
	https://doi.org/10.48550/arXiv.2510.18239 arXiv-issued DOI via DataCite

Submission history

From: Yihan Wu [view email]
[v1] Tue, 21 Oct 2025 02:53:17 UTC (9,042 KB)
[v2] Mon, 27 Oct 2025 21:18:47 UTC (9,042 KB)
[v3] Sat, 17 Jan 2026 19:53:33 UTC (9,038 KB)
[v4] Thu, 8 Oct 2026 17:45:35 UTC (9,246 KB)