arXiv cs.CL· Kilho Son, Paul Hsu, Cha Zhang, Dinei Florencio·· 12 小时前AI 评分31
RCEM：面向分布偏移的鲁棒对话式搜索嵌入模型
RCEM: Robust Conversational Search EMbedder in Distributional Shift
AI 导读
研究者提出 RCEM，一种结合 LLM 查询改写能力的鲁棒对话式搜索嵌入模型，通过将对话对齐到 LLM 改写后的查询而非长段落，在保持原嵌入空间不变的同时简化对齐任务。该方法无需对话到段落的相关性标注即可训练，并可直接复用原嵌入模型构建的索引而无需重建。实验显示 RCEM 在分布偏移下最多提升 30%，优于此前方法。
正文 · 原文
Abstract:We propose RCEM, a Robust Conversational search EMbedder that is additionally equipped with LLM's query reformulation capability without losing base model's generalization. Unlike prior conversational dense retrieval approaches that learn direct conversation-to-passage matching, RCEM aligns conversations, prepended by special token, to LLM-rewritten queries, while preserving the original embedding space. The unchanged embedding space automatically maps the rewritten-query to the relevant passages. As a result, RCEM (1) reduces overfitting by simplifying the alignment task from long passages to shorter rewritten queries, (2) eliminates the need for conversation-to-passage relevance labels for training, and (3) maintains its original embedding space that allows conversational queries against indexes built by original embedder without rebuilding them. Extensive experiments show that RCEM consistently outperforms prior approaches, achieving up to 30% improvement under distributional shift.
| Comments: | EMNLP |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.01697 [cs.CL] |
| (or arXiv:2606.01697v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.01697 arXiv-issued DOI via DataCite |
Submission history
From: Kilho Son [view email]
[v1]
Mon, 1 Jun 2026 05:04:17 UTC (516 KB)
[v2]
Wed, 17 Jun 2026 13:57:05 UTC (986 KB)
[v3]
Thu, 22 Oct 2026 23:34:32 UTC (935 KB)
来源：arXiv cs.CL · arxiv.org