New Information Retrieval (cs.IR) papers, Friday 23 October 2026

27 papers announced on arXiv. Get them ranked for you, free →

1. IRAH: Hierarchical Agentic Tree Search for Retrieval Pipeline Optimization

   Maria Ana Cardei, Aniket Singh, Amir Moghadam et al. · arXiv:2610.30167

   AI summaryIRAH jointly optimizes retrieval pipelines by hierarchical UCT-guided search (macro structure first, then parameters) with language-agent reflections and meta-learned warm starts. It outperforms TPE, achieving up to 6.4% nDCG@10 gains and the best average 0.4816 on four benchmarks.

2. FormBench: A Benchmark for Fine-Grained Retrieval in Formulation Patents

   Vineeth Venugopal, Elsa Olivetti, Jonathan Welch · arXiv:2610.30175

   AI summaryThe authors introduce FormBench, a benchmark with over 55,000 queries for evaluating fine-grained chemical retrieval in formulation patents. They show that commercial embedding APIs struggle significantly due to geometric collapse, while task-adapted open-source models and BM25 achieve the best retrieval performance.

3. FlexVDR: Learning to Allocate Matryoshka Representation Budgets for Visual Document Retrieval

   Haowen Xiang, Yibo Yan, Xin Zou et al. · arXiv:2610.30177

   AI summaryFlexVDR learns to assign each query a compact encoding route—varying depth, width, and token count—by predicting early‑stage utility, achieving near‑native retrieval quality while enabling configurable resource trade‑offs without retraining.

4. Association between Player Perception and AI Disclosure on Steam: An Exploratory Study

   Fabian Wiesinger, Johanna Pirker, Denis Helic · arXiv:2610.30178

   AI summaryThe study examines Steam reviews to see how AI disclosure affects player sentiment. It finds that games disclosing AI use receive more AI‑related criticism in negative reviews.

5. QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction

   Mianzhi Pan, Peishuo Liu, Yawen Ouyang et al. · arXiv:2610.30180

   AI summaryQProtFuse introduces a query-guided multi-layer feature fusion method for protein retrieval-augmented generation. It achieves state-of-the-art retrieval and function description performance by adaptively combining internal representations from protein language models.

6. Deep Search in Real-World Web Environments

   Xiaomin Yu, Hongliang Chen, Zhe Huang et al. · arXiv:2610.30181

   AI summaryDeep Search agents are evaluated on OpenDS, a live Web benchmark with 200 tasks requiring multi-step browsing, search, and evidence integration. The top model scores 39.0% accuracy, highlighting major challenges in tool coordination and long-horizon search.

7. Protection Is Not One Event: A Survey of RAG Protection Research by Measured Endpoint

   Peiyang Liu, Xi Wang, Di Liang et al. · arXiv:2610.30193

   Retrieval-augmented generation (RAG) systems face security, privacy, and copyright risks including content leakage, source reconstruction, unauthorized access, data poisoning, and prompt injection. Yet RAG protection studies use incompatible evaluation metrics, datasets, and…

8. FAIRGate: Agentic Runtime Policy Selection and Verified Repair for Fairer Recommendation Delivery

   Navkar Jain, Enock Ayiku, Kenneth Fletcher · arXiv:2610.30359

   AI summaryFAIRGate adds an agentic policy-selection layer that chooses from a library of fairness policies at runtime, applies constrained repair, and validates constraints before delivering top‑K rankings, enabling dynamic, measurable, and replayable fairness enforcement. Its main result is a model‑agnostic, runtime‑adjustable fairness controller that preserves relevance while satisfying multiple fairness constraints.

9. Data-Void Exploits in Web-Searching LLMs

   Michele Guerra, Wuyeh Jobe, Sanjana Nambiar et al. · arXiv:2610.30365

   AI summaryWeb‑searching LLMs can be manipulated by creating “data voids” where only attacker‑controlled pages exist; the study shows that a single such page appears in 31% of model responses across prompts and models, revealing a practical inference‑time attack vector.

10. When Agentic Retrieval Fails Silently: Framework and Production Evidence from a Creative AI Assistant

    Yan (Fergie), Liang, Anuj Agarwal et al. · arXiv:2610.30423

    AI summaryLayered Agentic Retrieval Assessment (LARA) measures retrieval quality, engagement, and outcome separately, addressing the attribution gap in agentic assistants. In Adobe Express, LARA restores retrieval attribution, showing chat users reformulate more often (44.4% vs. 33.1%) due to generated output refinement.

11. Getting a GRIP on Graph-Based Search for Learned Sparse Retrieval

    Vihan Lakshman, Franco Maria Nardini, Cosimo Rulli et al. · arXiv:2610.30425

    AI summaryGRIP improves graph‑based search for learned sparse retrieval by adding term‑based seeding, sparse routing, and partitioned construction, achieving 2.5× lower latency than Seismic with 95% recall@10 on MS MARCO. It shows graph indexes can compete with inverted indexes.

12. Journal Basket: Supporting Cross-Disciplinary Literature Searching and Methodological Transparency

    Michael Cahalane · arXiv:2610.30544

    Business students beginning a research project must learn how to identify relevant scholarly publications, distinguish disciplinary sources, construct database searches, and report the decisions behind their literature reviews. These tasks can be difficult because journal…

13. Better Reasoning, Not Just Bigger Models: A Structure-Grounded Framework for Efficient and Reliable Inference

    Amirhossein Aghaei, Liangyan Li, Shiyi Yang et al. · arXiv:2610.30611

    AI summarySGUR uses a training‑free graph‑matching and active query strategy to ground inference in structured hypothesis spaces, reducing uncertainty without large models. On a medical diagnosis task, SGUR achieves 92.5 % Hit@3, beating a Graph‑RAG baseline by 9.4 % while using five times fewer tokens.

14. The Fusion Tax: A Diagnostic Analysis of When Hybrid Retrieval Hurts

    Aaditya Chauhan, Arun Agarwal · arXiv:2610.30612

    AI summaryHybrid retrieval often hurts individual queries: the best fusion (dense‑weighted interpolation) places relevant docs below either component for 13–48% of queries and below dense‑only for 8–21%, even while winning aggregate recall. The paper recommends fusing only when the sparse‑dense balance is favorable.

15. Compiling Autoregressive LLM Control into Policy-Exact Parallel Inference for Hierarchical RAG

    Zehao Lin, Qianfeng Shen, Xixuan Hao et al. · arXiv:2610.30664

    AI summaryCompiles hierarchical RAG control into a policy‑exact parallel inference that replaces serial autoregressive decoding with a single root prefill and causal probes. It matches the serial policy on 84–89% of questions while cutting latency by 35–47% versus token‑wise decoding.

16. OneLatent: Latent Reasoning for Efficient Foundation Recommendation Models

    Hao Jiang, Peiru Du, Pengfei Yao et al. · arXiv:2610.30675

    AI summaryOneLatent compresses explicit chain‑of‑thought reasoning into learnable latent tokens, using multi‑view adaptive CoT and a curriculum‑based post‑training scheme. It improves Kuaishou recommendation metrics by ~17% while boosting inference throughput 17× and raising revenue 9.6% in production.

17. Lychee: Certified Screening of Inverted Lists for High-Recall Vector Search

    Dongfang Zhao · arXiv:2610.30697

    AI summaryLychee is a GPU-accelerated inverted-file vector search method that certifies and excludes non-qualifying vectors using tensor cores before computing exact distances. It achieves up to 117 times the speed of baseline methods at extreme recall levels on billion-scale datasets.

18. Automated Enrichment Analysis with GraphRAG: Efficient Multi-Database ORA Extraction and Biological Interpretation

    Guangcan Li, Shuyuan Zheng, Jiayao Liu et al. · arXiv:2610.30710

    AI summaryThe authors introduce Graph-Based Over-Representation Retrieval-Augmented Generation (GBOR-RAG), which models gene-set databases as a unified knowledge graph to streamline gene-expression analysis. The framework reduces query time by 37x while improving biological interpretation quality and claim faithfulness through compact evidence subgraphs.

19. Bezier-Guided Adaptive Flow for Cross-domain Sequential Recommendation

    Ziang Lu, Yu Zhang, Xinzhe Jiang et al. · arXiv:2610.30825

    AI summaryBGAFlow learns a shared semantic space and personalizes cross‑domain transfer by estimating a user‑specific horizon and guiding adaptation with a Bezier‑curve‑based flow that refines preference trajectories. It outperforms baselines on real datasets.

20. Lean Finder V2: Semantic Search for Mathlib That Disambiguates User Queries

    Jialin Lu, Xiaoyu Liu, Hengrui Luo et al. · arXiv:2610.30843

    AI summaryLean Finder V2 disambiguates short Mathlib queries using adversarial training and query rewriting, achieving 11 % better retrieval than previous engines on the MathlibQR benchmark and improving downstream theorem‑proving success when paired with Claude Code.

21. UserArena: Benchmarking Interactive User Simulation for Long-Horizon Shopping Agents

    Jujia Zhao, Jiakai Tang, Jing Yu et al. · arXiv:2610.30932

    AI summaryUserArena provides 64,729 real e‑commerce shopping sessions with an interactive simulator and offline validation against production A/B test results, enabling five‑level evaluation of agent policies. The proposed On‑Policy Trajectory Distillation method yields the highest simulator fidelity and best reproduces production policy effects.

22. PAIR: Preference-Aligned Intent Retrieval for Premium People Recommendation at LinkedIn

    Zhongkai Shangguan, Yanbin Kang, Yihan Cao et al. · arXiv:2610.30941

    General-purpose embeddings trained on profile similarity often misrank candidates in professional networks, where relevance is inherently complementary rather than symmetric. For example, job seekers usually engage with recruiters rather than peers with similar credentials.…

23. Formal-Informal Mathematical Alignment through and beyond Human Annotations

    Noah Gie{\ss}ing, Ankit Satpute, Hamid Rahkooy et al. · arXiv:2610.31011

    Formalized mathematics libraries such as Lean's \texttt{mathlib} provide machine-verifiable representations of mathematical knowledge, including definitions, theorems, and their relationships. If effectively connected to informal mathematical resources, formal mathematics could…

24. Chunk Recomputation under Real Document Revisions: Measuring Update Cost and Retrieval Quality of Fixed, Recursive, and Content-Defined Chunking

    Yuning Bian, Lele Xu, Yifan Yuan et al. · arXiv:2610.31019

    Chunking strategies for retrieval-augmented generation are almost always evaluated on a static snapshot, yet the documents underlying a knowledge base continue to change. Because chunks are embedded independently, a boundary that moves after an edit forces unchanged text to be…

25. The Delayed Countdown Method for Validating Review-Oriented AI Workflows in eDiscovery

    David D. Lewis, Tara Emory, Lenora Gray et al. · arXiv:2610.31311

    AI summaryThis paper introduces the Delayed Countdown method to validate AI-assisted document review workflows in eDiscovery. The approach easily integrates into existing systems and produces valid frequentist confidence intervals for review effectiveness regardless of the stopping rule used.

26. SurveyGraph: Generating Academic Surveys via Methods-Entity Knowledge Graphs

    Yifan Yuan, Yancheng Zhu, Zuoyu Zhang et al. · arXiv:2610.31317

    Most automated survey generation systems retrieve papers as flat documents, limiting their ability to capture research structure. We present SurveyGraph, a framework built around a Methods-Entity Knowledge Graph (MEKG) encoding method-level follow-on, competition, and…

27. A Task-Centric Survey of Long-Document Visual Question Answering: Taxonomy, Evolution, and Future Directions

    Yufei Tong, Guanjie Cheng, Tianle Wu et al. · arXiv:2610.31385

    Visual document understanding is essential for acquiring knowledge from visually rich documents and is attracting growing research interest. Existing surveys emphasize architectures or technical approaches, with limited attention to how task bottlenecks shape technical choices…