Document Embedding Preservation Tuning (DEPT) is introduced, which keeps tuned document embeddings close to cached initial embeddings while allowing retrieval gradients to pass through straight-through decoding into the generator.
Abstract
Large language models (LLMs) can both expand underspecified queries and encode text as dense representations, suggesting a unified model for query expansion and retrieval. Existing systems usually rely on prompted expansions, independently trained modules, or staged optimization, leaving generated expansions only indirectly aligned with the retrieval loss that judges them. We train a single decoder-only LLM end to end, where the same model generates the expansion and encodes both the expanded query and candidate documents. This unified setting creates a moving-target problem: retrieval supervision should improve query-side expansion, but the same update also shifts the document embeddings that serve as retrieval targets. We introduce Document Embedding Preservation Tuning (DEPT), which keeps tuned document embeddings close to cached initial embeddings while allowing retrieval gradients to pass through straight-through decoding into the generator. DEPT converts joint query--document movement into query-side adaptation against approximately stable, whitened document embeddings that support index reuse and online hard-negative mining. Experiments with Qwen3-4B-Instruct-2507 and LLaMA-3.2-3B-Instruct on five datasets in BEIR benchmark show that DEPT improves average retrieval quality over training-free, independently trained, and staged unified baselines, while ablations isolate the effects of preservation, whitening, end-to-end expansion training, and online negatives. Code is available at https://github.com/ILSparkle/DEPT.
This work introduces AnchorQE, a training-free method that separately encodes the original query and its expansion before interpolating them, and shows that AnchorQE improves retrieval effectiveness by up to 12.89% when compared to widely-used expansion-only or text-level concatenation baselines across TREC-DL, LoTTE, and BEIR.
Dense retrieval systems are core components of modern search engines, recommendation platforms, and retrieval-augmented generation pipelines. They encode documents and queries into dense embeddings, enabling efficient semantic search via vector similarity. However, because document embeddings are computed offline and stored in static indexes, these systems struggle to adapt to user feedback or evolving search intent. To address this limitation, we introduce \emph{embedding surgery}, a lightweight approach for adaptive ranking correction in dense retrieval. The method applies localized, minimal updates to selected document embeddings at query time, guided by editorial feedback, user interactions, or pseudo-labels from large language models. We formulate embedding surgery as a convex optimization problem that enforces ranking constraints while minimizing modifications to the affected document representations. We integrate embedding surgery into standard dense retrieval pipelines and evaluate it on TREC Deep Learning, TREC Robust, TREC CAsT, and MS MARCO benchmarks. Results show consistent improvements (e.g., up to +60.64\% relative improvement in nDCG@10 on DL-Hard under editorial feedback), even under noisy or shifting feedback, with low computational cost and without disrupting the global structure of the embedding space. Extensive experiments show that ranking corrections propagate to semantically related queries and that embedding updates can be applied safely and efficiently to scalable Approximate Nearest Neighbor indexes via simple in-place overwriting, without requiring costly index reconstruction. Finally, embedding surgery complements query adaptation methods such as CoRocchio, yielding additional gains while being more robust to noisy feedback.
Maddalena Amendola, Antonio Mallia, Raffaele Perego· 0 citations
This work presents DESA (Dense Expansion and Sparse Anchoring), which shares generated references across channels but specializes their integration, which improves nDCG@10 and Recall@20 over the unexpanded query and reduces dense and sparse replay stopping depths.
Large language models (LLMs) increasingly rely on information retrieval (IR) systems, such as Retrieval-Augmented Generation (RAG), to incorporate domain-specific knowledge without costly re-training. These systems often store pre-computed document embeddings in cloud-based vector databases. However, such embeddings are vulnerable to embedding inversion attacks (EIAs), which can reconstruct their underlying text. Existing defenses, such as adding noise or scaling embeddings, often provide limited privacy or significantly reduce retrieval utility. We propose SHAQ (shadow query generation), a semantic-decomposition and embedding-decoupling defense against EIAs. SHAQ is based on the insight that EIAs rely on the strong coupling between an embedding and its original text. Instead of storing document embeddings directly, SHAQ uses a generative language model to create diverse shadow queries that capture different semantic aspects of each document. These queries are then encoded and stored in place of the original document embeddings, thereby decomposing document semantics and decoupling stored embeddings from the source text. Experiments across diverse IR datasets show that SHAQ substantially improves privacy while preserving retrieval utility, achieving a recovery rate as low as 0.2104, defending up to 19.50% more tokens than baseline defenses, and reaching up to 0.7967 MAP@10 with up to 5.53% utility improvement. These results demonstrate that semantic decomposition and embedding decoupling provide an effective alternative to directly modifying embeddings for defending against EIAs.
Xinguo Feng, Zhongkui Ma, Zi-Han Wang et al.· 0 citations
Large Language Models (LLMs) have emerged as powerful assets for recommender systems. However, deploying them as generative recommenders or zero-shot rankers at web-scale remains bottlenecked by prohibitive computational overhead and grounding challenges. In this paper, we revitalize the classic, highly efficient two-tower retrieval architecture by adapting LLMs as semantic representation backbones rather than generative engines. We introduce an LLM-native two-tower framework engineered for high-throughput, large-scale retrieval. Our architecture introduces several key innovations: a shared LLM encoder for joint user-item modeling, End-Of-Sentence (EOS) token pooling for compact sequence embedding, cross-dataset transfer learning, knowledge distillation from powerful cross-encoder teachers, and latent reasoning within the user tower. Extensive evaluation across three public benchmarks demonstrates that cross-encoder architecture outperforms current state-of-the-art (SoTA) models, while the efficient two-tower student achieves SoTA-comparable retrieval performance. Furthermore, experiments on internal large-scale production systems yield substantial topline retrieval improvements along with high resilience to model staleness and superior data scaling. Our findings demonstrate that when augmented with modern representation learning, the traditional two-tower paradigm remains an exceptionally competitive and practical solution for industrial retrieval systems.
Document Visual Question Answering (DocVQA) often leverages Retrieval-Augmented Generation (RAG), where late-interaction encoders are commonly used to identify document pages relevant to a user query, before answer generation by a Large Vision-Language Model (LVLM). Existing approaches typically retrieve a fixed top-$k$ number of pages regardless of query complexity, which increases LVLM latency and may degrade answer accuracy. We introduce ViSAR (Visual Semantic Activation Retrieval), a training-free adaptive-$k$ retrieval method for late-interaction visual document retrieval. ViSAR operates directly in the embedding space to construct a query-conditioned page-level similarity matrix that highlights query-relevant semantics and dynamically determines the number of pages to retrieve. Across multiple encoders and LVLMs, ViSAR retrieves compact, query-adapted page sets that reduce RAG latency by up to 58.7\%, while maintaining or improving answer accuracy compared with fixed top-$k$ and adaptive retrieval heuristics. Furthermore, we show that the similarity matrix structure correlates with answer accuracy, suggesting future directions for retrieval quality-aware document understanding.
Adrien Mialland, Marc Plantevit, Julien Gallois et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
