Papers

Large language model-based recommender systems (LLM-RSs) have demonstrated remarkable capabilities, but are computationally unsustainable for many real-world applications. Compact LLMs offer a practical alternative, yet their reduced capacity often requires reasoning or knowledge distillation methods that increase latency or depend on larger models. Combined with autoregressive generation, these approaches face severe scalability bottlenecks. In contrast, discriminative LLM-RSs enable efficient full-corpus ranking through embedding similarity, but compact backbones remain limited in expressiveness and structural adaptivity. We propose the Fusion of Layer-wise Exits for Sequential Recommendation (FLEXRec), a discriminative framework that enhances compact LLMs while retaining scalable full-corpus ranking. FLEXRec inserts prediction heads (i.e., exits) at multiple transformer layers and adaptively fuses their score distributions. An adaptive continuous router (AC-Router) dynamically selects both the number and identity of exits for each user sequence, while a novel target-k hinge loss regulates routing sparsity. Experiments on three real-world datasets with Qwen 3 1.7B and Llama 3.2 3B show that FLEXRec achieves state-of-the-art accuracy among competing methods while remaining highly efficient. Code: https://github.com/xurong-liang/FLEXRec

Semi-structured documents are ubiquitous in scientific reports, financial statements, and technical manuals. Question answering over such documents requires simultaneous understanding of text, tables, charts, and complex hierarchical layouts. Existing methods either rely on repeatedly calling large language models for structure parsing and retrieval, leading to high cost and large latency, or they flatten the document and lose layout and hierarchy information, sacrificing answer accuracy. To address this, we propose EidosDoc, a novel system that achieves state-of-the-art accuracy with minimal computational expense. Our approach introduces three core innovations. (1) An Implicit Structure Encoder trained via contrastive learning and a structure consistency loss. This module jointly embeds hierarchical relationships, spatial positions, and textual content into a dense vector space, capturing document structure holistically without the need for manually defined and error-prone constructions. (2) A Hybrid Retrieval Pipeline that leverages BM25, layout fingerprints, and a lightweight cross-encoder to perform high-precision retrieval entirely without invoking an LLM, drastically reducing cost and latency. (3) A Dynamic Evidence Expansion mechanism that adaptively retrieves spatially adjacent and structurally related evidence, overcoming the evidence omission common in fixed-path retrieval methods. We evaluate EidosDoc on four benchmarks, and comprehensive evaluations show that EidosDoc achieves a new state-of-the-art accuracy on the four benchmarks. Crucially, it does so with a 50 times reduction in cost and 4 times lower latency compared to the previous state-of-the-art Method. These results demonstrate that EidosDoc establishes a new optimal trade-off among accuracy, cost, and speed, offering a practical and scalable path for accurate semi-structured document analysis.

In the past few years, consumers have moved the bulk of their product exploration and purchasing efforts online seeking speed, convenience, and price comparison with ease unimaginable for in-person shopping. As product catalogs have grown in diversity and size product search and recommendation have become a cornerstone for e-commerce sites. Despite the widespread usage of search engines in e-commerce, there is no high-quality dataset designed to evaluate end-to-end retrieval quality. In 2025, we ran a revised and continued version of the Product Search track previously run at TREC 2023 and TREC 2024. The 2025 product search track had two tasks: query expansion and related-product recommendation. The related-product recommendation task is particularly novel, providing an annotated data set of product relationships that distinguishes between complementary and related products. We anticipate the data from this track will enable better recommendation and search applications that reflect user needs, as a building block for conversational product discovery experiences.

Offline evaluation is a major gateway before online evaluation of ranking models in A/B testing. Standard offline metrics measure predictive accuracy, but are only a surrogate for downstream utility: a model can improve them while redistributing impressions across objective buckets in ways that degrade downstream utility. No offline method surfaces these impression share shifts before online evaluation. We propose impression share prediction as an offline evaluation task: given a candidate ranking model, predict the distribution of impressions it would produce across objective buckets - impressions grouped by optimization goal (e.g., click, video view). The task is inherently counterfactual, since the candidate has never served live traffic. We propose a structural causal model of how model predictions and delivery capacity jointly determine impression allocation, and show the counterfactual effect is identified from observational data. Building on this, we develop a statistical learning framework that predicts impression shares from a candidate's early-interaction confidence signals and current system state, trained on historical data. On data from multiple ranking model families, a Random Forest reduces L1 error by 49% over a constant baseline for models seen during training. For held-out models, evaluated by time since first appearance, the first hour is the closest analog to true online evaluation and the hardest: the Random Forest falls below the baseline because the capacity state still reflects the prior model. An encoder-conditioned architecture that simulates a 2-hour rollout over recent auction dynamics recovers $+$22% L1 in this regime.

Generative Engine Optimization (GEO) modifies web content to increase its likelihood of being selected and cited by generative search engines. This can give strategically optimized pages visibility disproportionate to their authority or relevance and even make weak or false information appear well supported. Unlike conventional search, generative search synthesizes information into direct answers rather than presenting competing sources, which can further amplify these risks, as assessing source provenance and authority requires additional user interaction. Despite these concerns, systematic methods for detecting GEO-optimized webpages remain underexplored. We introduce GEOFlagBench, a benchmark of 3,200 web content instances spanning 400 queries, four domains, and eight GEO optimizer families, and use it to systematically evaluate existing GEO detection methods. Although the strongest baseline achieves an aggregate F1 of 0.880, method-level and authorship-conditioned evaluations reveal substantial weaknesses and potential reliance on authorship-related shortcuts. We therefore propose Intervention-Paired Training (IPT), which supervises detector responses to GEO interventions and non-GEO AI polishing; on ModernBERT, IPT improves F1 from 0.862 to 0.944 and worst-group accuracy from 0.725 to 0.883. We develop a GEO-gated Agent system for auditing the Source Tier and verifiability of Citation URLs in detected GEO pages. Finally, we deploy the complete pipeline on released Google Search and Gemini-grounded retrieval results for 1,000 real-user queries. Across 10,095 available pages, we estimate an overall GEO prevalence of 8.90%, reaching 16.36% among pages modified in 2026. Our results establish a foundation for systematically detecting, auditing, and measuring GEO in real-world search ecosystems.

Industrial recommenders rely on two model families that have evolved largely independently: feature-interaction models over multi-field user/item features, and sequential models over user-behavior histories. Production systems couple them only loosely. To unify the two, we present UniDot, a novel architecture for post-click conversion prediction built from the factorization-machine (FM) point of view: the embedding inner product—which powers collaborative filtering and lets a recommender generalize to unseen user–item pairs—is the same primitive as attention's query dot key scoring, so a single dot-product of tokens can underlie both feature interaction and sequence modeling. UniDot tokenizes non-sequential fields and multi-domain behavioral sequences into one shared token space and stacks a single macro-block in which a token-mixing bus and a sequence-retrieval bus (item tokens cross-attending the histories) run in parallel and exchange state each layer through an MLP-Mixer fusion, while an FM Highway carries explicit per-layer dot-product interactions around the residual stack directly to the classifier. The sequence side is embedded once per forward pass and shared by all consumers, bounding inference latency. Trained with a dual sparse/dense (Adagrad + Muon) optimizer, an auxiliary conversion-delay head, and multi-path mutual learning, UniDot finished as the runner-up on the Industrial track of the TAAC KDD Cup 2026.

Recommender systems typically rely on implicit feedback (e.g., clicks) to infer user preferences. However, such data is inherently prone to various biases, including position bias and popularity bias. Position bias occurs when higher-ranked items receive more interactions regardless of true relevance. Popularity bias reinforces frequent exposure of popular items while under-recommending relevant, yet less popular ones. Directly learning from such data fails to capture true user preferences, leading to suboptimal recommendations. This research focuses on mitigating position bias and popularity bias in recommender systems. Specifically, I address position bias in learning-to-rank (LTR) systems and popularity bias in collaborative filtering (CF) models and social recommender systems based on graph neural networks. My work develops methods that overcome the limitations of existing approaches to mitigating position bias and popularity bias, enabling more relevant and personalized recommendations that align with users' preferences.

Retrieval-augmented and agentic question-answering systems increasingly re-derive the meaning of a corpus at query time. Put plainly, instead of re-deriving what a corpus means on every question, the work is done once when a document arrives and is thereafter merely consulted – a compiler, not an interpreter, of meaning. An alternative is to compile that meaning once, at ingest time, into a compact, queryable semantic substrate and maintain it as the corpus evolves. The central objection is maintenance cost: rebuilding a truncated singular value decomposition (SVD) on every change appears prohibitive, and a change of embedding model seems to force a full re-embedding. We argue and show empirically that maintenance cost scales with the amount of change, not corpus size. On a controlled synthetic pilot (dimension 256, rank 32, a corpus grown from 3,000 to 9,000 documents over 50 update events), incremental low-rank updates were 33.7 times cheaper per update than full re-SVD and 23.8 times cheaper cumulatively, while the incremental subspace tracked the full recomputation to within floating-point precision (maximum principal-angle drift below 1e-11 degrees; recall@10 = 1.0). An orthogonal Procrustes virtual axis update recovered 0.95 mean cosine to truly re-embedded vectors by re-embedding only about 10 percent of the corpus. The results support maintaining, rather than repeatedly reconstructing, a semantic substrate.

Negative sampling is a key component of implicit collaborative filtering (CF), as it enables recommenders to effectively learn user preferences. Existing negative sampling methods mostly follow a two-stage paradigm: they first construct a candidate negative pool for each user and then select negative samples from the pool according to predefined sampling rules. However, these methods usually overlook the hardness variation of candidate negative pools across users, making it difficult to adaptively adjust the hardness and informativeness of negative samples according to candidate-pool conditions. In addition, most existing samplers evaluate candidate negatives mainly through a matching score computed from the final aggregated user and item embeddings, while ignoring the structural differences captured by multi-hop neighborhood aggregation. As a result, the training value of negatives may be insufficiently characterized. To address these issues, we propose SAHC-NS, a Structure-Aware and Hardness-Calibrated Negative Sampling method. Specifically, SAHC-NS uses the mean and standard deviation of layer-wise matching scores to capture the overall matching strength and cross-layer structural discrepancy of candidate negatives, respectively. This enables SAHC-NS to select informative negatives by taking cross-layer structural discrepancy into account, rather than relying solely on final matching scores. Moreover, SAHC-NS introduces a candidate-pool-aware hardness calibration module to dynamically adjust negative augmentation strength according to candidate-pool hardness, producing hardness-controllable negatives. Extensive experiments demonstrate the superiority of SAHC-NS over existing negative sampling methods.

Dense retrieval is commonly evaluated on benchmarks that represent each document with a single embedding, even though real-world retrieval systems often index long documents that require chunking. In these settings, the chosen chunking method not only affects retrieval quality, but also indexing throughput, query latency, and memory usage. Prior comparisons of chunking strategies have mainly focused on retrieval performance, leaving operational trade-offs underexplored. To address these issues, we evaluate eight representative chunking strategies across two scalable corpora, three embedding models, and multiple corpus sizes, measuring both retrieval effectiveness and system-level costs. Our results show that computationally expensive methods rarely provide consistent gains over simpler chunking. Instead, the best performing strategy depends on the embedding model, dataset, corpus size, and target retrieval metric. Methods with similar performance can also differ substantially in operational cost, showing that chunking should be seen as a multi-objective design decision.

Large-scale agents increasingly rely on retrieval to access external capabilities. We study this retrieval gate in structured tools and APIs, a measurable class of tool-backed executable skills that must be surfaced before an agent can plan, incorporate, or act. In this setting the retrieval layer can silently fail even when the capability corpus is fixed: on ToolRet, a retriever fine-tuned on one source-specific slice collapses on another source-specific slice of the same benchmark, with FT-1100 despite its higher lexical overlap with the gold tools. We call this failure mode source-style collapse. Query-side TF-IDF fingerprints flag source styles on which the fine-tuned retriever is likely to fail better than semantic or length-based proxies, giving a cheap signal for mismatch over a fixed tool corpus. We propose ToolScout, a source-aware routing method that uses this signal as a routing guard: on the mixed 4,996-query stream, TF-IDF-based routing raises coverage from 22.3% to 86.1%, and across five collapsed sources 20 matched examples raise the coverage-weighted global top-1 proxy from 1.3% to 53.9%. The same failure and routing behaviors persist when tools are rerendered as executable skill cards, which rules out raw API-schema format as the sole cause.

Range-filtering approximate nearest neighbor search (RFANNS) is a fundamental operation in modern vector databases. Given a query vector $q$ and a numerical range predicate, RFANNS returns the $k$-approximate nearest neighbors ($k$-ANN) of the query $q$ among the objects whose attributes satisfy the range predicate. However, existing RFANNS methods are not well suited to high-throughput GPU execution. CPU indexes offer limited parallel scalability, generic GPU filtering is highly selectivity-dependent, and GPU indexes built from locally optimized subgraphs can incur long search trajectories and redundant distance computations. To address these limitations, we present FROG, a GPU-oriented RFANNS index that replaces multiple locally optimal substructure building with a globally aware, vertex-centric design. It organizes diverse expansion neighbor candidates for each vertex in a GPU-friendly structure and rapidly identifies the expansion neighbors used for computation at query time. Moreover, GPU-oriented algorithms and implementations are developed for both index construction and query processing. Experiments on six datasets show that FROG improves mixed-selectivity query throughput by 14.7–37.7$\times$ over 44-core CPU baselines and 4.5–7.6$\times$ over the strongest GPU baseline. It also accelerates index construction by 2.4–14.8$\times$ over the GPU baseline.

Range-filtered approximate nearest neighbor search (RFANNS) is an important primitive for vector databases; it retrieves vectors that are similar to a query and satisfy a numerical range predicate, but existing RFANNS indexes expose vectors, attributes, and queries in plaintext. This assumption is unsuitable for outsourced vector databases, where sensitive data and queries must be protected from an honest-but-curious cloud server. To the best of our knowledge, this is the first study that systematically formulates and evaluates privacy-preserving RFANNS over outsourced encrypted vector databases. Our approach separates range localization from encrypted vector search: an authorized user maps the query range to a compact set of nodes in a local N-ary attribute tree, and the server searches only the corresponding proximity graph sub-indices over encrypted vectors. To reduce expensive encrypted comparisons, we use a filter-and-refine pipeline that first retrieves coarse candidates with approximate distance-comparison-preserving encryption and then reranks a small candidate set with exact distance-comparison encryption. We then analyze the computation, storage, communication, and leakage of the protocol. Experiments on four widely used vector datasets show that our method improves the QPS-Recall trade-off over representative secure adaptations of existing RFANNS approaches, scaling effectively to large datasets.

Research software forms distinct co-usage communities that span traditional disciplinary boundaries, yet the structure of these communities remains largely unexplored. We present a graph-based framework for discovering mathematical software communities and predicting their association with research publications. We construct a software co-usage network from publication-software relationships using a curated swMATH dataset and subsequently apply community detection method, revealing a heterogeneous landscape of mathematical software communities. We formulate publication-to-community mapping as a multi-label classification task and further investigate whether community membership can be predicted from lightweight scholarly metadata. Specifically, we compare two feature representations of scientific publications: Mathematics Subject Classification (MSC) and title-based embeddings. Across a range of models, structured MSC representation consistently provides a stronger precision-recall trade-off, demonstrating that structured domain metadata captures software-community structure more effectively than compressed title-only semantics in this setting. This work highlights the continuing value of structured scholarly metadata for large-scale research software discovery, classification and recommendation.

Point-of-interest (POI) recommendation models based on graph neural networks achieve strong performance by propagating collaborative signals over user-item interactions, yet they struggle with the cold-start problem, where items with few or no interactions are not represented. In this paper, we propose LLM-augmented Multi-Graph Contrastive Learning (LLM-MGCL), a multi-graph neural network that uses semantic and spatial information about items to extend the LightGCN backbone with two auxiliary item-item graphs: a semantic graph constructed from sentence embeddings of LLM-generated photo summaries and keywords, and a geographic graph derived from Haversine distances between business locations. Item embeddings are propagated over all three graphs in parallel, fused additively, and aligned across views through a bidirectional InfoNCE contrastive objective that connects behavioral, semantic, and spatial representations of the same items. Experiments on the Yelp Multimodal Recommendation Dataset show that LLM-MGCL outperforms classical collaborative filtering, matrix factorization, and interaction-only graph neural network baselines. It improves Recall@20 by 52.0% and NDCG@20 by 64.8% over LightGCN while performing on par with the strongest contrastive baseline, Self-supervised Graph Learning (SGL), which is also affected by the cold-start problem. An ablation study reveals that the cross-view contrastive alignment (CA) is the primary driver of these gains, with the best performance achieved when all three graphs are combined. Our results suggest that externally grounded, LLM-derived item knowledge can effectively compensate for missing collaborative signal and mitigate the item cold-start problem in POI recommendation.

Generating regulation-compliant test scenarios is essential for validating safety-critical automotive systems, yet Large Language Models (LLMs) struggle to ground outputs in long, hierarchical standards. We present RegulaRAG, a Retrieval-Augmented Generation (RAG) pipeline that couples SmartChunking, reference-aware enrichment of paragraphs and tables via graph traversal, with Smart Retrieve & Rerank over these enriched units. To test our system, we evaluate on a manually curated dataset covering all scenarios in UN Regulation No. 152 (AEBS). Our study comprises: (i) a three-step progressive search that identifies near-optimal retrieval parameters without exhaustive grid search; (ii) head-to-head comparisons against five baseline RAG systems; and (iii) a robustness stress test that scales the source corpus with distractor content. Outputs are evaluated using a customized penalized scoring metric. Across all experiments, RegulaRAG achieves the highest average Meta-Score (82.99), outperforming the next-best system by 43% (NoRAG: 57.94), while operating at 14k-25k tokens per query versus up to 500k for graphcentric baselines. It maintains strong performance, remaining stable even as the number of regulatory sources grows, whereas competing RAG systems degrade sharply in both quality and robustness.

Static pruning is widely used to accelerate sparse neural retrieval, yet existing studies each validate their conclusions within a single custom pipeline, leaving it unclear which findings transfer to modern engines with different index organizations and dynamic pruning mechanisms. We present the first cross-engine pruning portability study, evaluating static pruning strategies across three engines - a controlled C++ pipeline (exhaustive inverted index), BMP (block-max pruning), and SEISMIC (clustered inverted indexes) - on two benchmarks (MS MARCO, Natural Questions) with two encoders spanning opposite query-density regimes (SPLADE: 44 avg. query terms; V3-GTE: 7 avg. query terms), totaling 1,140 experimental configurations, with an additional deep-judgment validation on TREC DL 2019/2020. We find that index-side pruning (document and posting-list) is portable: it consistently reduces latency (1.2-6.6$\times$) and index size (18-82%) across all engines because sparse retrieval is memory-bound - a conclusion we support with cache-miss, TLB, and IPC profiling. In contrast, query pruning is already internalized by modern engines: it yields 4-11$\times$ speedup on the exhaustive pipeline but is subsumed by BMP's $\beta$ and SEISMIC's query_cut. Static pruning complements dynamic pruning: on BMP, combining document and query reduction yields 2.5$\times$ speedup with NDCG@10 within 0.003 of the exact baseline. Finally, NDCG@10 saturates while Recall@10 is still in the ${\sim}$85-95% range across all three engines, providing a portable stopping criterion: practitioners can push pruning to this knee without visible ranking degradation. Together, these findings answer what transfers (index-side pruning), what breaks (query pruning), and what still helps (static atop dynamic pruning).

Positional encoding is a fundamental component of Transformer-based generative recommendation models, where user histories are modeled as autoregressive item sequences. Most positional encoding methods are inherited from natural language processing and mainly represent discrete item order. However, recommendation sequences go beyond ordered lists, as timestamps and temporal effects also shape item relations. Our work is motivated by a real-world food delivery and instant retail recommendation system, where user behavior exhibits multi-level temporal regularities, including recency effects, meal-time peaks, weekday-weekend shifts, and promotion-driven traffic bursts. Existing methods partially address this issue through timestamp features, interval embeddings, decay functions, or attention biases, but they usually inject heterogeneous temporal signals through a unified representation or a single modeling pathway, making it difficult to distinguish broad temporal dynamics from local order cues. To address this limitation, we propose Decoupled Temporal Encoding (DTE), a lightweight framework for generative recommendation. DTE separates temporal dynamics from order information through two complementary modules: a personalized macro-temporal module that injects compact temporal primitives into item embeddings, and a time-gated micro-sequential module that introduces relative-order bias only when interactions are temporally dense. DTE is also parameter-efficient and deployment-friendly, allowing easy integration into existing systems.

General-purpose text embedding models are designed to capture semantic similarity but are not optimised for distinguishing entity records that represent the same real-world business or person. This limitation affects applications such as entity resolution and duplicate record retrieval, where small textual differences may either preserve or change identity. This paper investigates whether domain-specific triplet fine-tuning can adapt pretrained embedding models for identity-sensitive retrieval. A synthetic dataset of business and person records was created with identity-preserving variations and challenging non-matching examples. Two widely used embedding models were evaluated before and after fine-tuning using a margin-based similarity evaluation. The results show substantial improvements in separating true matches from highly similar non-matches, demonstrating that domain-specific triplet training can effectively reshape general-purpose embedding spaces for entity retrieval. These findings suggest that targeted fine-tuning provides a practical approach for improving embedding models in data quality management and information retrieval applications.

Enterprises connect language models to their own data through retrieval. The benchmarks that rank multi-hop retrieval systems leave out two facts a buyer needs before a published number can be used: whether the retrieval backbone may be deployed commercially, and what it costs to build. On licensing: the field's dense-retrieval anchor, NV-Embed-v2, is licensed cc-by-nc-4.0. Of the four leading MuSiQue systems we audit (HippoRAG-2, PropRAG, SAG, KET-RAG), three depend on it for their best numbers and none says so. On performance: we measure thirteen embedders from eight makers on one identical MuSiQue harness with bootstrap confidence intervals throughout. Until mid-2026 there was a real commercial tax: the best commercially-licensed embedder trailed the anchor by 2.31 Recall@5 points (95% CI [0.91, 3.71], p=0.001). NVIDIA's Nemotron-3-Embed-8B, released 2026-07-16, has closed it: +0.24 at Recall@5 (95% CI [-0.94, +1.43], p=0.69), -0.58 at Recall@10 (p=0.28). It matches the anchor, does not beat it, and is the only entrant that is commercially licensed, free to self-host, and indistinguishable from the anchor; every other entrant meeting the first two conditions sits 5.2 to 14.6 points below. The durable finding is the paid-versus-free divide: API embedders charge per token on every re-index, self-hosted ones charge nothing. On cost: three of five audited systems (adding Microsoft's GraphRAG) do not disclose indexing cost, and the only published GraphRAG dollar figures span 11x inside one third-party paper (USD 2.30 vs USD 24.94 to index a 5.64 MB corpus once); extrapolated to 1 TB that undisclosed choice separates roughly USD 428K from $4.6M. Our cost model keeps one-time embedding apart from recurring answering: at 1 TB, embedding sits 7.5x-900x below graph construction, and a year of answering at 10,000 queries/day sits 350x or more below it.

Continual recommendation aims to capture evolving user interests from streaming data but struggles with sparsity. LLM enhancers mitigate this with semantic knowledge, but naive integration creates a new conflict. We identify this as the Stability-Plasticity-Cognitivity (SPC) Trilemma, where generalized LLM semantic priors (Cognitivity) conflict with retaining personalized historical preferences (Stability) and adapting to individual interest shifts (Plasticity). To address this, we propose Trilemma-Responsive Adaptive Continual Enhancement for Recommendation (TRACER). TRACER synergistically combines three specialized modules, each targeting stability, plasticity, or cognitivity, while preventing any single lemma from dominating. This holistic design enables semantic knowledge to support history retention and adaptation to evolving interests without disrupting continual learning. Across five real-world datasets, TRACER effectively harmonizes the SPC trilemma and outperforms state-of-the-art baselines by up to 14.38%. Our code is available at https://github.com/woo-joo/TRACER_CIKM26.

Sequential recommenders often struggle with sparse and noisy histories, limiting generalization to unseen interactions. Knowledge distillation mitigates this by transferring dense supervision from a teacher to a student. However, most distillation methods run teacher and student independently, then match student outputs or representations to the teacher. Such supervision entangles student-component effects, blurring whether weak generalization stems from unreliable embeddings, overfitted encoding, or co-adaptation to sparse histories. In this paper, we propose Graft-Oriented Distillation (GOD), a component-level distillation framework for improved generalization through grafting. Grafting denotes replacing selected frozen-teacher components with trainable student counterparts to build hybrid source models. GOD uses these hybrid models to evaluate student embeddings with the teacher encoder and the student encoder with teacher embeddings, providing component-level feedback. At inference, GOD uses only the student, incurring no additional cost. Across three real-world datasets, GOD outperforms state-of-the-art baselines by up to 13.92%.

Pseudo-query generation can alleviate the supervision bottleneck for agent skill retrieval, but existing document-level approaches typically leave the rich internal relations among capabilities, parameters, and usage examples implicit. As a result, generated queries may be topically relevant to a skill while lacking capability grounding and parameter consistency, raising the question of whether explicitly exploiting a skill document's internal structure can produce more effective retrieval signals. We therefore propose Skill2Query, a framework that first parses a skill document into a Skill Knowledge Graph and then generates pseudo-queries through a three-stage process including style mimicking, query template generation, and parameter filling. The generated queries can be used for offline index augmentation, online query expansion, and retriever training. Four benchmarks (TheoremQA, LogicBench, ToolQA, and CHAMP) are used to evaluate Skill2Query with large-scale skill candidate pools across multiple downstream applications, including skill retrieval, retriever training, and end-to-end agent execution. Using nearly 30K skills across diverse domains, we generate 700K category-diverse pseudo-queries. Skill2Query consistently improves sparse, dense, and skill-routing retrieval, with an average Recall@1 gain of 6.70 percentage points across retrieval settings. Skill2Query-generated training data also achieves the best Recall@1 and nDCG@1 among the evaluated generation baselines. Further evaluations with multiple LLM backends demonstrate that improved skill retrieval translates into higher agent task success rates. Code and resources are available at https://github.com/MatZaharia/Skill2Query.

Spreadsheets are a primary medium for publishing tabular data, yet automatically extracting structured content from them remains difficult due to heterogeneous layouts, diverse file formats, and inconsistent organizational conventions. We address two core tasks in spreadsheet understanding: Cell-Type Classification (CTC), which assigns roles to cells, and Table Detection (TD), which identifies table bounding boxes within sheets. We propose an efficient two-stage pipeline in which a learned CTC model feeds a deterministic TD algorithm. For CTC, we use a LightGBM classifier over 65 structured features together with a pairwise CRF enforcing spatial consistency across the cell grid. Our TD method extracts table ranges from predicted cell types by a deterministic five-stage procedure. For evaluation, we built and share StatSheets, a multilingual benchmark of 737 manually annotated sheets from 14 public data providers across multiple countries and file formats. Under 5-fold cross-validation, our CRF-LightGBM system achieves a Mean File-Macro F1 score of 0.937 on CTC, within 0.6 percentage points of the GPU-based TUTA Transformer, while requiring substantially fewer computational resources. For TD, our deterministic approach outperforms region-based baselines and remains competitive with recent LLM-based systems such as SpreadsheetLLM. These results demonstrate that combining non-linear structured prediction with deterministic range extraction provides a competitive, scalable, and computationally efficient approach to spreadsheet table understanding.

Retrieval-augmented generation (RAG) answers a question by retrieving passages from a vector store and trusting them as context, so anyone who can add documents can try to steer the answer. A recent, appealing defense filters poisoning at ingestion, rejecting any document that behaves like a hub. We show it – and every ingestion-time filter – is defeated by a coordinated adversary that injects a handful of individually unremarkable documents which together surround one target query and seize its top-k (on BGE-large / BEIR, m=10 documents take 10/10; 9.9/10 on a live HNSW index). The attack is not theoretical. Realized as ordinary fluent text and run end-to-end through a BGE-large + HNSW + Qwen2.5-7B pipeline, it makes the generator emit the attacker's planted claim in 88% of targets, versus 0% without the injection. And no admission-time defense stops it: at ingestion an attack cone is geometrically identical to a legitimate niche upload, so – measuring this directly – the strongest trained classifier, given every feature and thousands of examples, separates the two no better than chance, catching 4.2% of attacks at a 1% false-positive rate. We prove this limit for the entire class of ingestion-time statistics (any decision from documents and reference queries alone), and it reproduces – and worsens – across two corpora and five encoders. The one signal that separates an attack from legitimate niche ingestion – a query's demand – is invisible before retrieval, which is also the escape: a retrieval-time detector that observes demand catches 100% of the attacks at the same 1% false-positive rate. Coverage of the query space by an admission gate is not containment of coordinated poisoning; robust defense must move past the front door, to demand.

In a vector database serving production-scale retrieval, a single inserted document can be retrieved for an anomalously large share of the query workload – a retrieval hub – and dominate the evidence returned for an entire topic. An emerging defense guards against this at ingest with an admission check: it maintains a set of sentinel queries and admits a document only if its reverse-kNN count against them stays below a threshold tau. Under workload drift this sentinel set is a query-aware auxiliary index that must be maintained online, and we study the cost that maintenance imposes on the ingest path. We identify a structural limit – coverage is not redundancy: a monitor stops promoting sentinels once a region is covered, but the predicate rejects a hub only once tau sentinels witness it, so exposure has an observation-limited floor that no reduction in update or enforcement latency can close. On real HNSW, IVF-Flat, and IVF-PQ indexes over an 8.8M-vector MS MARCO corpus this floor is only a best case: as index recall falls, exposure and churn rise above it, and below recall ~0.5 the gate stops containing altogether – worst on the memory-compressed IVF-PQ used at billion scale – while a recall-aware witness probe restores containment at a fixed O(|S|d) admission cost, under 0.1% of the ANN insert. We validate the law under real (COVID-19) workload drift, implement the gate in PostgreSQL/pgvector at a 0.33% ingest tax, and turn the bound into a provisioning rule that sizes the sentinel budget per emerging region. A count test contains the hub where retrieval-time score normalizers (NNN, QB-Norm) do not, and a pre-registered causal suite isolates the missing-coverage mechanism from retrieval fragmentation across two embedding families (BGE-1024, E5-768).

Graph-based Retrieval-Augmented Generation (GraphRAG) retrieves evidence for multi-hop questions over structured cor- pus graphs. Existing GraphRAG methods leave the connection between evidence discovery and source grounding implicit. We propose LineageRAG, which constructs one evidence lin- eage for each query-derived evidence demand and completes it with a verbatim source span when the selected evidence supports that demand. LineageRAG first initializes the evi- dence demands. It then expands each lineage through demand- conditioned retrieval over the corpus graph while retaining the demand associated with every candidate. Lineage completion uses this provenance to select complementary passages and grounds supported demands in verbatim source text. Experi- ments on HotpotQA, 2WikiMultiHopQA, and MuSiQue show that LineageRAG improves R@5, EM, and F1 by 3.51, 5.96, and 5.22 points on average over leading GraphRAG baselines.

Sparse long-context inference requires efficient token retrieval in both prefill and decode. Existing methods often use different retrieval strategies for the two stages, preventing one retrieval representation from being reused throughout inference. We propose Self-Indexing Attention, a training-free framework built on a shared transform-domain sign-magnitude representation. The key signs provide a reusable token-level index for grouped prefill selection and decode retrieval, while the same representation remains compatible with external KV-cache compression without separate indexer metadata. This 1-bit index enables efficient retrieval through bitwise operations widely supported by modern accelerators. At 5% attention density, Self-Indexing Attention remains close to dense attention on LongBench and RULER and achieves up to 6.1x prefill and 10.3x decode attention-operator speedups. Experiments with TurboQuant and DeepSeekV4-Flash further demonstrate compatibility with low-bit KV-cache compression and pretrained sparse-attention indexers.

Recent advances in large language models (LLMs) have enabled their use as conversational recommender systems (CRS), demonstrating strong recommendation accuracy and natural dialogue. However, guiding multi-turn interactions to elicit user preferences effectively remains challenging. Existing approaches either use separate reinforcement learning agents with templated interactions or optimize for interactivity judged by another LLM, without measuring how much useful information is actually gained. We propose a new approach that quantifies the effectiveness of each interaction by the reduction in the assistant's uncertainty, measured via entropy over recommendations. We apply this entropy reduction as a reward—without relying on ground-truth recommendations, which are often unavailable in real-world scenarios—to fine-tune the LLM, enabling strategic interaction generation. Empirical results with supervised fine-tuning (SFT) and direct preference optimization (DPO) on the INSPIRED and ReDial datasets show that our method improves both recommendation quality and conversational efficiency.

Retrieval-Augmented Generation over knowledge graphs (Graph-RAG) has emerged as a powerful paradigm for grounding large language models in domain-specific corpora. However, existing systems face persistent limitations: (1) static chunking fragments long documents, losing cross-section semantic connections; (2) ingestion pipelines do not scale adaptively; and (3) multi-domain deployments require either a monolithic knowledge base that dilutes retrieval precision or manual user routing. We present Noesis, a decoupled Graph-RAG architecture addressing these limitations through four algorithms: (a) Bidirectional Graph Traversal with a Graph-Feedback Context Resolver simulating human reading with degrading memory; (b) an AIMD Concurrency Controller adapted from TCP congestion control, achieving 23x speedup with zero OOM events; (c) Moesis, domain-aware selective quantization for MoE models achieving 6.3x speedup on 12 GB consumer GPUs; and (d) Mesh, cross-KB semantic routing with runtime structural discovery enabling small on-premises models to perform multi-hop cross-domain reasoning. On HotpotQA (1,000 questions), Noesis achieves 59.5 EM / 74.7 F1, surpassing GraphRAG by +27.8 EM while using a 35B on-premises model for graph construction rather than GPT-4o. Source text verification on a 193-page document confirms 90% precision on long-range causal edges inaccessible to chunk-independent extraction.