Dense Retrieval

Large language model (LLM) based listwise reranking has emerged as the dominant paradigm for achieving state-of-the-art ranking effectiveness in information retrieval. However, its reliance on feeding full passage texts into the LLM introduces two critical bottlenecks: the "lost in the middle" phenomenon degrades ranking quality as input length grows, and the inference latency scales super-linearly with sequence length, rendering it impractical for industrial deployment. In this paper, we present ResRank, a unified retrieval-reranking framework that fundamentally addresses both challenges. Inspired by multimodal LLMs that project visual inputs into compact token representations, ResRank employs an Encoder-LLM to compress each candidate passage into a single embedding, which is then fed alongside the query text into a Reranker-LLM for listwise ranking. To alleviate the misalignment between the compressed representation space and the ranking space, we introduce a residual connection structure that combines encoder embeddings with contextualized hidden states from the reranker. Furthermore, we replace the conventional autoregressive decoding with a one-step cosine-similarity-based scoring mechanism, eliminating the generation bottleneck entirely. ResRank is trained through a carefully designed dual-stage, multi-task, end-to-end joint optimization strategy that simultaneously trains the encoder and reranker, achieving learning objective alignment between retrieval and reranking while substantially reducing training complexity. Extensive experiments on TREC Deep Learning and eight BEIR benchmark datasets demonstrate that ResRank achieves competitive or superior ranking effectiveness compared to existing approaches while requiring zero generated tokens and processing only one token per passage, yielding a fundamentally better balance between effectiveness and efficiency.

Modern retrieval-augmented generation (RAG) systems treat vector embeddings as static, context-free artifacts: an embedding has no notion of when it was created, how trustworthy its source is, or which other embeddings depend on it. This flattening of knowledge has a measurable cost: recent work on VersionRAG reports that conventional RAG achieves only 58% accuracy on versioned technical queries, because retrieval returns semantically similar but temporally invalid content. We propose SmartVector, a framework that augments dense embeddings with three explicit properties -- temporal awareness, confidence decay, and relational awareness -- and a five-stage lifecycle modeled on hippocampal-neocortical memory consolidation. A retrieval pipeline replaces pure cosine similarity with a four-signal score that mixes semantic relevance, temporal validity, live confidence, and graph-relational importance. A background consolidation agent detects contradictions, builds dependency edges, and propagates updates along those edges as graph-neural-network-style messages. Confidence is governed by a closed-form function combining an Ebbinghaus-style exponential decay, user-feedback reconsolidation, and logarithmic access reinforcement. We formalize the model, relate it to temporal knowledge graph embedding, agentic memory architectures, and uncertainty-aware RAG, and present a reference implementation. On a reproducible synthetic versioned-policy benchmark of 258 vectors and 138 queries, SmartVector roughly doubles top-1 accuracy over plain cosine RAG (62.0% vs. 31.0% on a held-out split), drops stale-answer rate from 35.0% to 13.3%, cuts Expected Calibration Error by nearly 2x (0.244 vs. 0.470), reduces re-embedding cost per single-word edit by 77%, and is robust across contradiction-injection rates from 0% to 75%.

While Large Language Models (LLMs) exhibit exceptional zero-shot relevance modeling, their high computational cost necessitates framing passage retrieval as a budget-constrained global optimization problem. Existing approaches passively rely on first-stage dense retrievers, which leads to two limitations: (1) failing to retrieve relevant passages in semantically distinct clusters, and (2) failing to propagate relevance signals to the broader corpus. To address these limitations, we propose Bayesian Active Learning with Gaussian Processes guided by LLM relevance scoring (BAGEL), a novel framework that propagates sparse LLM relevance signals across the embedding space to guide global exploration. BAGEL models the multimodal relevance distribution across the entire embedding space with a query-specific Gaussian Process (GP) based on LLM relevance scores. Subsequently, it iteratively selects passages for scoring by strategically balancing the exploitation of high-confidence regions with the exploration of uncertain areas. Extensive experiments across four benchmark datasets and two LLM backbones demonstrate that BAGEL effectively explores and captures complex relevance distributions and outperforms LLM reranking methods under the same LLM budget on all four datasets.

Retrieval-Augmented Generation (RAG) has become a standard approach for enhancing large language models (LLMs) with external knowledge, mitigating hallucinations, and improving factuality. However, existing systems rely on generating natural language queries at each hop and maintaining a strict architectural separation between retriever and generator, preventing them from leveraging the full representational capacity of the LLM. We propose LAnR (Latent Abstraction for RAG), a unified framework in which a single LLM jointly performs encoding, retrieval, and generation entirely within its own latent space. Rather than generating textual queries, LAnR produces dense retrieval vectors from the hidden states of a designated [PRED] token and uses them to match against encoded document representations from the same model. Furthermore, LAnR adaptively decides when sufficient evidence has been retrieved using a lightweight MLP control head over those same hidden states, eliminating both the separate retriever and explicit token-level stopping reasoning. This design is motivated by our empirical observation that answer token entropy reliably signals retrieval sufficiency. Extensive experiments on six QA benchmarks spanning single-hop and multi-hop settings demonstrate that LAnR outperforms existing RAG methods, while achieving improved inference efficiency through reduced number of retrieval calls and tighter model integration.

Retrieval-Augmented Generation (RAG) systems depend on the geometric properties of vector representations to retrieve contextually appropriate evidence. When source documents interleave multiple topics within contiguous text, standard vectorization produces embedding spaces in which semantically distinct content occupies overlapping neighborhoods. We term this condition semantic entanglement. We formalize entanglement as a model-relative measure of cross-topic overlap in embedding space and define an Entanglement Index (EI) as a quantitative proxy. We argue that higher EI constrains attainable Top-K retrieval precision under cosine similarity retrieval. To address this, we introduce the Semantic Disentanglement Pipeline (SDP), a four-stage preprocessing framework that restructures documents prior to embedding. We further propose context-conditioned preprocessing, in which document structure is shaped by patterns of operational use, and a continuous feedback mechanism that adapts document structure based on agent performance. We evaluate SDP on a real-world enterprise healthcare knowledge base comprising over 2,000 documents across approximately 25 sub-domains. Top-K retrieval precision improves from approximately 32% under fixed-token chunking to approximately 82% under SDP, while mean EI decreases from 0.71 to 0.14. We do not claim that entanglement fully explains RAG failure, but that it captures a distinct preprocessing failure mode that downstream optimization cannot reliably correct once encoded into the vector space.

Decoder-only large language models (LLMs) are increasingly replacing BERT-style architectures as the backbone for dense retrieval, achieving substantial performance gains and broad adoption. However, the robustness of these LLM-based retrievers remains underexplored. In this paper, we present the first systematic study of the robustness of state-of-the-art open-source LLM-based dense retrievers from two complementary perspectives: generalizability and stability. For generalizability, we evaluate retrieval effectiveness across four benchmarks spanning 30 datasets, using linear mixed-effects models to estimate marginal mean performance and disentangle intrinsic model capability from dataset heterogeneity. Our analysis reveals that while instruction-tuned models generally excel, those optimized for complex reasoning often suffer a ``specialization tax,'' exhibiting limited generalizability in broader contexts. For stability, we assess model resilience against both unintentional query variations~(e.g., paraphrasing, typos) and malicious adversarial attacks~(e.g., corpus poisoning). We find that LLM-based retrievers show improved robustness against typos and corpus poisoning compared to encoder-only baselines, yet remain vulnerable to semantic perturbations like synonymizing. Further analysis shows that embedding geometry (e.g., angular uniformity) provides predictive signals for lexical stability and suggests that scaling model size generally improves robustness. These findings inform future robustness-aware retriever design and principled benchmarking. Our code is publicly available at https://github.com/liyongkang123/Robust_LLM_Retriever_Eval.

Standard dense retrievers lack a native calculus for multi-atom logical constraints. We introduce Neuro-Symbolic Fuzzy Logic (NSFL), a framework that adapts formal t-norms and t-conorms to neural embedding spaces without requiring retraining. NSFL operates as a first-order hybrid calculus: it anchors logical operations on isolated zero-order similarity scores while actively steering representations using Neuro-Symbolic Deltas (NS-Delta) -- the first-order marginal differences derived from contextual fusion. This preserves pure atomic meaning while capturing domain reliance, preventing the representation collapse and manifold escape endemic to traditional geometric baselines. For scalable real-time retrieval, Spherical Query Optimization (SQO) leverages Riemannian optimization to project these fuzzy formulas into manifold-stable query vectors. Validated across six distinct encoder configurations and two modalities (including zero-shot and SOTA fine-tuned models), NSFL yields mAP improvements up to +81%. Notably, NSFL provides an additive 20% average and up to 47% boost even when applied to encoders explicitly fine-tuned for logical reasoning. By establishing a training-free, order-aware calculus for high-dimensional spaces, this framework lays the foundation for future dynamic scaling and learned manifold logic.

Large language models (LLMs) have demonstrated strong capabilities in medical question answering; however, purely parametric models often suffer from knowledge gaps and limited factual grounding. Retrieval-augmented generation (RAG) addresses this limitation by integrating external knowledge retrieval into the reasoning process. Despite increasing interest in RAG-based medical systems, the impact of individual retrieval components on performance remains insufficiently understood. This study presents a systematic evaluation of retrieval-augmented medical question answering using the MedQA USMLE benchmark and a structured textbook-based knowledge corpus. We analyze the interaction between language models, embedding models, retrieval strategies, query reformulation, and cross-encoder reranking within a unified experimental framework comprising forty configurations. Results show that retrieval augmentation significantly improves zero-shot medical question answering performance. The best-performing configuration was dense retrieval with query reformulation and reranking achieved 60.49% accuracy. Domain-specialized language models were also found to better utilize retrieved medical evidence than general-purpose models. The analysis further reveals a clear tradeoff between retrieval effectiveness and computational cost, with simpler dense retrieval configurations providing strong performance while maintaining higher throughput. All experiments were conducted on a single consumer-grade GPU, demonstrating that systematic evaluation of retrieval-augmented medical QA systems can be performed under modest computational resources.

Multimodal retrieval systems struggle to resolve image-text queries against text-only corpora: the best vision-language encoder achieves only 27.6 nDCG@10 on MM-BRIGHT, underperforming strong text-only retrievers. We argue the bottleneck is not the retriever but the query -- raw multimodal queries entangle visual descriptions, conversational noise, and retrieval intent in ways that systematically degrade embedding similarity. We present BRIDGE, a two-component system that resolves this mismatch without multimodal encoders. FORGE (Focused Retrieval Query Generator) is a query alignment model trained via reinforcement learning, which distills noisy multimodal queries into compact, retrieval-optimized search strings. LENS (Language-Enhanced Neural Search) is a reasoning-enhanced dense retriever fine-tuned on reasoning-intensive retrieval data to handle the intent-rich queries FORGE produces. Evaluated on MM-BRIGHT (2,803 queries, 29 domains), BRIDGE achieves 29.7 nDCG@10, surpassing all multimodal encoder baselines including Nomic-Vision (27.6). When FORGE is applied as a plug-and-play aligner on top of Nomic-Vision, the combined system reaches 33.3 nDCG@10 -- exceeding the best text-only retriever (32.2) -- demonstrating that query alignment is the key bottleneck in multimodal-to-text retrieval. https://github.com/mm-bright/multimodal-reasoning-retrieval

Retrieval shapes how language models access and ground knowledge in retrieval-augmented generation (RAG). In historical research, the target is often not an arbitrary relevant passage, but the exact record for a specific regnal month, where temporal consistency matters as much as topical relevance. This is especially challenging for Classical Chinese annals, where time is expressed through terse, implicit, non-Gregorian reign phrases that must be interpreted from surrounding context, so semantically plausible evidence can still be temporally invalid. We introduce ChunQiuTR, a time-keyed retrieval benchmark built from the Spring and Autumn Annals and its exegetical tradition. ChunQiuTR organizes records by month-level reign keys and includes chrono-near confounders that mirror realistic retrieval failures. We further propose CTD (Calendrical Temporal Dual-encoder), a time-aware dual-encoder that combines Fourier-based absolute calendrical context with relative offset biasing. Experiments show consistent gains over strong semantic dual-encoder baselines under time-keyed evaluation, supporting retrieval-time temporal consistency as a key prerequisite for faithful downstream historical RAG. Our code and datasets are available at github.com/xbdxwyh/ChunQiuTR.

Dense retrieval systems increasingly face an asymmetry between complex instruction-like queries and relatively simple, static document collections. While stronger embedders can better understand such queries, re-embedding large corpora or fine-tuning large models is often impractical. We propose Efficient Retrieval Adapter (ERA), a query-side adapter learning framework for re-index-free retrieval adaptation. ERA first aligns the embedding spaces of a strong query embedder and a lightweight document embedder using unlabeled corpus documents, and then adapts the aligned query representation with a small number of labeled query-document pairs. Across 126 MAIR retrieval tasks from six domains, ERA improves average nDCG@10 by up to 8.2 points in symmetric settings and by more than 12 points in asymmetric settings, while using substantially fewer labels than supervised adapter training. These results show that retrieval systems can benefit from stronger query understanding without updating backbone embedders or rebuilding document indexes.

Downlink bottlenecks motivate onboard systems that prioritize hazards without transmitting raw pixels. We study a strict setting where a ground station uplinks only compact embeddings plus metadata, and an onboard system performs vector search to triage new captures. We ask whether this embedding-only pipeline remains useful under explicit remote-sensing shift: cross-time (pre/post-event), cross-event/location (different disasters), cross-site cloud (15 geographic sites), and cross-city AOI holdout (buildings). Using OlmoEarth embeddings on a scaled public multi-task benchmark (27 Sentinel-2 L2A scenes, 15 cloud sites, 5 SpaceNet-2 AOIs; 10 seeds), we find that all effective methods rely on the same uplinked embeddings, but the optimal decision head is task-dependent: kNN retrieval is significantly superior for cloud classification (0.92 vs. centroid 0.91; p<0.01, Wilcoxon), while class centroids dominate temporal change detection (0.85 vs. retrieval 0.48; p<0.01). These results show that embedding-only uplink is the key enabler--once embeddings are onboard, the system can select the best head per task at no additional uplink cost, with all telemetry under 1 KB per query.

As AI agents become the primary consumers of retrieval APIs, there is an opportunity to expose more of the retrieval pipeline to the caller. flexvec is a retrieval kernel that exposes the embedding matrix and score array as a programmable surface, allowing arithmetic operations on both before selection. We refer to composing operations on this surface at query time as Programmatic Embedding Modulation (PEM). This paper describes a set of such operations and integrates them into a SQL interface via a query materializer that facilitates composable query primitives. On a production corpus of 240,000 chunks, three composed modulations execute in 19 ms end-to-end on a desktop CPU without approximate indexing. At one million chunks, the same operations execute in 82 ms.

Which properties of hard-negative supervision track dense-retriever transfer before training? We compare lexical noncoverage, frozenencoder similarity statistics, hardness proxies, and combinations of these signals across BM25, dense, LLM-generated, and hybrid sources for MS MARCO training. Evaluation uses three retriever backbones and eleven BEIR datasets. Mean lexical residuality, an IDF-weighted measure of query-token noncoverage, reproduces the aggregate transfer ordering in all six backbone and setting blocks. The individual semantic statistics recover none of the complete orderings, and combining them with the lexical signal does not improve the observed rank agreement. After approximately matching BM25, dense, and LLM diagnostic samples for lexical residuality, none of the tested semantic statistics recovers the observed LLM > Dense > BM25 transfer ordering across three backbones, three matching resolutions, and 100 repetitions. Constructed controls further show that unrelated and false-negative-shaped passages can exhibit favorable lexical and semantic statistics. The findings identify a simple lexical association in this setting while distinguishing source level transfer agreement from the reliability of individual negatives.

Dimensionality reduction is critical for deploying dense retrieval systems at scale, yet mainstream post-hoc methods face a fundamental trade-off: principal component analysis (PCA) preserves dominant variance but underutilizes representational capacity, while whitening enforces isotropy at the cost of amplifying noise in the heavy-tailed eigenspectrum of retrieval embeddings. Intermediate spectral scaling methods unify these extremes by reweighting dimensions with a power coefficient $γ$, but treat $γ$ as a fixed hyperparameter that requires task-specific tuning. We show that the optimal scaling strength $γ$ is not a global constant: it varies systematically with target dimensionality $k$ and is governed by the signal-to-noise ratio (SNR) of the retained subspace. Based on this insight, we propose Spectral Tempering (SpecTemp), a learning-free method that derives an adaptive $γ(k)$ directly from the corpus eigenspectrum using local SNR analysis and knee-point normalization, requiring no labeled data or validation-based search. Extensive experiments demonstrate that Spectral Tempering consistently achieves near-oracle performance relative to grid-searched $γ^*(k)$ while remaining fully learning-free and model-agnostic. Our code is publicly available at https://github.com/liyongkang123/SpecTemp.

Domain-specific finetuning is essential for dense retrievers, yet not all data pairs contribute equally to the learning process. We introduce OPERA, a data pruning framework that exploits this heterogeneity to improve both the effectiveness and efficiency of retrieval model adaptation. We first investigate static pruning (SP), which retains only high-similarity query-document pairs, revealing an intrinsic quality-coverage tradeoff: ranking (NDCG) improves while retrieval (Recall) can degrade due to reduced query diversity. To resolve this tradeoff, we propose a two-stage dynamic pruning (DP) strategy that adaptively modulates sampling probabilities at both query and document levels throughout training, prioritizing high-quality examples while maintaining access to the full training set. Evaluations across eight datasets spanning six domains demonstrate the effectiveness of both approaches: SP improves ranking over standard finetuning (NDCG@10 +0.2 points), while DP achieves the strongest performance on both ranking (NDCG@10 +1.0 points) and retrieval (Recall@20 +0.4 points), with an average rank of 1.38 across all methods. These findings scale to Qwen3-Embedding, an LLM-based dense retriever, confirming architecture-agnostic benefits. Notably, DP reaches comparable performance in less than 50% of the training time required by standard finetuning.

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.

Recent advances in Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) have enabled diverse retrieval methods. However, existing retrieval methods often fail to accurately retrieve results for negation and exclusion queries. To address this limitation, prior approaches rely on embedding adaptation or fine-tuning, which introduce additional computational cost and deployment complexity. We propose Direct Embedding Optimization (DEO), a training-free method for negation-aware text and multimodal retrieval. DEO decomposes queries into positive and negative components and optimizes the query embedding with a contrastive objective. Without additional training data or model updates, DEO outperforms baselines on NegConstraint, with gains of +0.0738 nDCG@10 and +0.1028 MAP@100, while improving Recall@5 by +6% over OpenAI CLIP in multimodal retrieval. These results demonstrate the practicality of DEO for negation- and exclusion-aware retrieval in real-world settings.

Maximum inner product search (MIPS) is a crucial subroutine in machine learning, requiring the identification of a vector taken within a database (the keys) that best aligns with a given query. We propose amortized MIPS: a regression-based approach that trains neural networks to directly predict MIPS solutions, amortizing the cost of repeatedly solving MIPS for queries drawn from a known distribution over a fixed key database. Our key insight is that the MIPS value function is the support function of the set of keys, a well-studied convex function whose gradient yields the optimal key. This motivates two complementary amortized models: SupportNet, an input-convex neural network trained to regress the support function, and KeyNet, a vector-valued network that directly regresses the optimal key. SupportNet can serve as a cluster router, steering queries toward relevant database partitions, while KeyNet can be used as a drop-in replacement for the original query, fed directly to off-the-shelf indexing pipelines. Our experiments on the BEIR benchmark show that, for document embeddings, learned \SupportNet{}s and \KeyNet{}s significantly improve IVF match rates when accounting for compute effort, whether measured in FLOPs, number of probes, or wall-clock time. Our code is available at: https://github.com/apple/ml-amips.

We present the first large-scale, cross-domain evaluation of document chunking strategies for dense retrieval, addressing a critical but underexplored aspect of retrieval-augmented systems. In our study, 36 segmentation methods spanning fixed-size, semantic, structure-aware, hierarchical, adaptive, and LLM-assisted approaches are benchmarked across six diverse knowledge domains using five different embedding models. Retrieval performance is assessed using graded relevance scores from a state-of-the-art LLM evaluator, with Normalised DCG@5 as the primary metric (complemented by Hit@5 and MRR). Our experiments show that content-aware chunking significantly improves retrieval effectiveness over naive fixed-length splitting. The top-performing strategy, Paragraph Group Chunking, achieved the highest overall accuracy (mean nDCG@50.459) and substantially better top-rank hit rates (Precision@124%, Hit@559%). In contrast, simple fixed-size character chunking as baselines performed poorly (nDCG@5 < 0.244, Precision@12-3%). We observe pronounced domain-specific differences: dynamic token sizing is strongest in biology, physics and health, while paragraph grouping is strongest in legal and maths. Larger embedding models yield higher absolute scores but remain sensitive to suboptimal segmentation, indicating that better chunking and large embeddings provide complementary benefits. In addition to accuracy gains, we quantify the efficiency trade-offs of advanced chunking. Producing more, smaller chunks can increase index size and latency. Consequently, we identify methods (like dynamic chunking) that approach an optimal balance of effectiveness and efficiency. These findings establish chunking as a vital lever for improving retrieval performance and reliability.

Deep Research agents are rapidly emerging as primary consumers of modern retrieval systems. Unlike human users who issue and refine queries without documenting their intermediate thought processes, Deep Research agents generate explicit natural language reasoning before each search call, revealing rich intent and contextual information that existing retrievers entirely ignore. To exploit this overlooked signal, we introduce: (1) Reasoning-Aware Retrieval, a retrieval paradigm that jointly embeds the agent's reasoning trace alongside its query; and (2) DR-Synth, a data synthesis method that generates Deep Research retriever training data from standard QA datasets. We demonstrate that both components are independently effective, and their combination yields a trained embedding model, AgentIR-4B, with substantial gains. On the challenging BrowseComp-Plus benchmark, AgentIR-4B achieves 68% accuracy with the open-weight agent Tongyi-DeepResearch, compared to 50% with conventional embedding models twice its size, and 37% with BM25. Code and data are available at: https://texttron.github.io/AgentIR/.

Nepali, a low-resource language, faces significant challenges in building an effective information retrieval system due to the unavailability of annotated data and computational linguistic resources. In this study, we attempt to address this gap by preparing a pair-structured Nepali Question-Answer dataset. We focus on Frequently Asked Questions (FAQs) for passport-related services, building a data set for training and evaluation of IR models. In our study, we have fine-tuned transformer-based embedding models for semantic similarity in question-answer retrieval. The fine-tuned models were compared with the baseline BM25. In addition, we implement a hybrid retrieval approach, integrating fine-tuned models with BM25, and evaluate the performance of the hybrid retrieval. Our results show that the fine-tuned SBERT-based models outperform BM25, whereas multilingual E5 embedding-based models achieve the highest retrieval performance among all evaluated models.

Prompt-only, single-step LLM query rewriting, where a rewrite is generated from the query alone without retrieval feedback, is commonly used in production RAG pipelines, but its effect on dense retrieval is poorly understood. We present a systematic empirical study across three BEIR benchmarks, two dense retrievers, and multiple training configurations, and find strongly domain-dependent behavior: rewriting degrades nDCG@10 by 9.0 percent on FiQA, improves it by 5.1 percent on TREC-COVID, and has no significant effect on SciFact. We identify a consistent mechanism: degradations co-occur with reduced lexical alignment between rewritten queries and relevant documents, as rewriting replaces domain-specific terms in already well-matched queries. In contrast, improvements arise when rewriting shifts queries toward corpus-preferred terminology and resolves inconsistent nomenclature. Lexical substitution occurs in 95 percent of rewrites across all outcome groups, showing that effectiveness depends on the direction of substitution rather than substitution itself. We also study selective rewriting and find that simple feature-based gating can reduce worst-case regressions but does not reliably outperform never rewriting, with even oracle selection offering only modest gains. Overall, these results show that prompt-only rewriting can be harmful in well-optimized verticals and suggest that domain-adaptive post-training is a safer strategy when supervision or implicit feedback is available.

LLMs have fundamentally transformed dense retrieval, upgrading backbones from discriminative encoders to generative architectures. However, a critical disconnect remains: while LLMs possess strong reasoning capabilities, current retrievers predominantly utilize them as static encoders, leaving their potential for complex reasoning unexplored. To address this, existing approaches typically adopt rewrite-then-retrieve pipelines to generate explicit CoT rationales before retrieval. However, this incurs prohibitive latency. In this paper, we propose LaSER, a novel self-distillation framework that internalizes explicit reasoning into the latent space of dense retrievers. Operating on a shared LLM backbone, LaSER introduces a dual-view training mechanism: an Explicit view that explicitly encodes ground-truth reasoning paths, and a Latent view that performs implicit latent thinking. To bridge the gap between these views, we design a multi-grained alignment strategy. Beyond standard output alignment, we introduce a trajectory alignment mechanism that synchronizes the intermediate latent states of the latent path with the semantic progression of the explicit reasoning segments. This allows the retriever to think silently and effectively without autoregressive text generation. Extensive experiments on both in-domain and out-of-domain reasoning-intensive benchmarks demonstrate that LaSER significantly outperforms state-of-the-art baselines. Furthermore, analyses across diverse backbones and model scales validate the robustness of our approach, confirming that our unified learning framework is essential for eliciting effective latent thinking. Our method successfully combines the reasoning depth of explicit CoT pipelines with the inference efficiency of standard dense retrievers.

The de facto paradigm for applying dense retrieval (DR) to new tasks involves fine-tuning a pre-trained model for a specific task. However, this paradigm has two significant limitations: (1) It is difficult adapt the DR to a new domain if the training dataset is limited. (2) Old DR models are simply replaced by newer models that are trained from scratch when the former are no longer up to date. Especially for scenarios where the model needs to be updated frequently, this paradigm is prohibitively expensive. To address these challenges, we propose a novel dense retrieval approach, termed dynamic dense retrieval (DDR). DDR uses prefix tuning as a module specialized for a specific domain. These modules can then be compositional combined with a dynamic routing strategy, enabling highly flexible domain adaptation in the retrieval part. Extensive evaluation on six zero-shot downstream tasks demonstrates that this approach can surpass DR while utilizing only 2% of the training parameters, paving the way to achieve more flexible dense retrieval in IR. We see it as a promising future direction for applying dense retrieval to various tasks.

Embedding-based retrieval (EBR) for large-scale e-commerce search faces three intertwined challenges: graded (non-binary) relevance where engagement signals are noisy and intent-varying while business relevance guidelines admit acceptable-but-not-exact matches, false negatives in hard sample mining, and unstable similarity score separability across relevance levels, the last of which complicates hybrid search score fusion and downstream ranking. We propose Mine and Refine, a two-stage contrastive training framework that addresses all three. A lightweight LLM, fine-tuned with engagement-driven audit, serves as a guideline-aligned scalable labeler throughout training. Stage 1 establishes a robust global embedding space via label-aware supervised contrastive learning; Stage 2 mines hard samples, re-annotates them with the LLM labeler to mitigate spurious negatives, and refines the model through a multi-level extension of circle loss that enforces margin-controlled separation across relevance levels. Deployed in production e-commerce search across multiple product verticals, the approach delivers statistically significant lifts in user engagement and gross order value, and substantially improves retrieval and end-to-end relevance metrics.

Long-tail question answering presents significant challenges for large language models (LLMs) due to their limited ability to acquire and accurately recall less common knowledge. Retrieval-augmented generation (RAG) systems have shown great promise in mitigating this limitation by integrating external retrieval mechanisms. However, dense retrieval models often face the same difficulties when generalizing to rare or niche knowledge. In this study, we introduce RPDR, a novel data augmentation framework that selects high-quality easy-to-learn training data, to enhance dense retrievers. Our approach is built around three core components: synthetic data generation, data selection with Round-Trip prediction to identify easy-to-learn instances, and retriever training with these instances. We evaluate RPDR on two long-tail retrieval benchmarks, PopQA and EntityQuestion, demonstrating substantial improvements over existing retrievers like BM25 and Contriver, especially on extremely long-tail categories. We identify the strengths and limitations of RPDR through detailed human analysis and propose a dynamic routing mechanism to dynamically route queries to specialized retrieval modules to further improve retrieval performance.

We introduce WebFAQ 2.0, a new version of the WebFAQ dataset, containing 198 million FAQ-based natural question-answer pairs across 108 languages. Compared to the previous version, it significantly expands multilingual coverage and the number of bilingual aligned QA pairs to over 14.3M, making it the largest FAQ-based resource. Unlike the original release, WebFAQ 2.0 uses a novel data collection strategy that directly crawls and extracts relevant web content, resulting in a substantially more diverse and multilingual dataset with richer context through page titles and descriptions. In response to community feedback, we also release a hard negatives dataset for training dense retrievers, with 1.25M queries across 20 languages. These hard negatives were mined using a two-stage retrieval pipeline and include cross-encoder scores for 200 negatives per query. We further show how this resource enables two primary fine-tuning strategies for dense retrievers: Contrastive Learning with MultipleNegativesRanking loss, and Knowledge Distillation with MarginMSE loss. WebFAQ 2.0 is not a static resource but part of a long-term effort. Since late 2025, structured FAQs are being regularly released through the Open Web Index, enabling continuous expansion and refinement. We publish the datasets and training scripts to facilitate further research in multilingual and cross-lingual IR. The dataset itself and all related resources are publicly available on GitHub and HuggingFace.

Reliable retrieval-augmented generation (RAG) systems depend fundamentally on the retriever's ability to find relevant information. We show that neural retrievers used in RAG systems have blind spots, which we define as the failure to retrieve entities that are relevant to the query, but have low similarity to the query embedding. We investigate the training-induced biases that cause such blind spot entities to be mapped to inaccessible parts of the embedding space, resulting in low retrievability. Using a large-scale dataset constructed from Wikidata relations and first paragraphs of Wikipedia, and our proposed Retrieval Probability Score (RPS), we show that blind spot risk in standard retrievers (e.g., CONTRIEVER, REASONIR) can be predicted pre-index from entity embedding geometry, avoiding expensive retrieval evaluations. To address these blind spots, we introduce ARGUS, a pipeline that enables the retrievability of high-risk (low-RPS) entities through targeted document augmentation from a knowledge base (KB), first paragraphs of Wikipedia, in our case. Extensive experiments on BRIGHT, IMPLIRET, and RAR-B show that ARGUS achieves consistent improvements across all evaluated retrievers (averaging +3.4 nDCG@5 and +4.5 nDCG@10 absolute points), with substantially larger gains in challenging subsets. These results establish that preemptively remedying blind spots is critical for building robust and trustworthy RAG systems.

Adapting general-purpose LLMs into domain-specialized dense retrievers typically requires large-scale training on mixed-domain data. We show that merging independently trained domain-specialized experts consistently exceeds this approach across four decoder-only LLM families (0.6B-7B), four merging methods, and twelve medical and general retrieval tasks from MTEB, suggesting that parameter-space composition captures complementary domain strengths that large-scale mixed-domain training averages out. To further maximize expert quality, we introduce Synthesize-Train-Merge (STM), a modular framework that synthesizes hard negatives with a top-tier LLM and fine-tunes domain-specialized experts via LoRA before merging them, without continual pre-training. Synthesized hard negatives yield the largest gains for smaller models, and STM achieves strong performance on biomedical retrieval tasks while maintaining competitive general-domain results across all four backbone families.