让检索模型更懂语义组合，提升对细微语义变化的敏感度。
Training for Compositional Sensitivity Reduces Dense Retrieval Generalization
- 用结构化负样本训练，增强模型对语义组合变化的感知能力。
- 在零样本测试中，小模型性能下降8-9%，大模型下降高达40%。
- 适合关注文本检索鲁棒性与语义理解的NLP研究者。
密集检索将文本压缩为单一嵌入，通过余弦相似度排序。尽管召回效率高，但在身份匹配任务上表现脆弱：微小的组合修改（如否定、角色互换）会改变语义却保持高相似度。受单位球面余弦空间几何性质启发（Kang et al., 2025），我们检验了纯文本检索中的这一矛盾。在四个双编码器骨干网络上，加入结构化负样本可稳定降低零样本NanoBEIR检索性能（小模型平均nDCG@10下降8-9%，大模型最高达40%），但对整体空间分离改善有限。将池化余弦视为召回接口后，我们评估了基于词元-词元余弦图的验证器。MaxSim（后期交互）在重排序中表现优异，但无法拒绝结构近似错误；而一个小规模Transformer在端到端训练下能可靠区分此类近似错误。
原文摘要 · Abstract (English)
Dense retrieval compresses texts into single embeddings ranked by cosine similarity. While efficient for recall, this interface is brittle for identity-level matching: minimal compositional edits (negation, role swaps) flip meaning yet retain high similarity. Motivated by geometric results for unit-sphere cosine spaces (Kang et al., 2025), we test this retrieval-composition tension in text-only retrieval. Across four dual-encoder backbones, adding structure-targeted negatives consistently reduces zero-shot NanoBEIR retrieval (8-9% mean nDCG@10 drop on small backbones; up to 40% on medium ones), while only partially improving pooled-space separation. Treating pooled cosine as a recall interface, we then benchmark verifiers scoring token--token cosine maps. MaxSim (late interaction) excels at reranking but fails to reject structural near-misses, whereas a small Transformer over similarity maps reliably separates near-misses under end-to-end training.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。