My reading, by research topic

1 · Reasoning-first ranking & LLM rerankers

ACL A Survey of Reasoning-Intensive Retrieval: Progress and Challenges — Systematizes reasoning-intensive retrieval benchmarks and methods by where and how reasoning enters the pipeline. DOI arXiv
ACL Adaptive Retrieval for Reasoning — Studies when to retrieve during reasoning, proposing adaptive retrieval for large reasoning models. DOI
ACL Contextual Relevance and Adaptive Sampling for LLM-Based Document Reranking — TS-SetRank, an uncertainty-aware sampling reranker built on contextual relevance, lifts nDCG@10 by 15-25% on BRIGHT and 6-21% on BEIR. arXiv
ACL Learning from Emptiness: De-biasing Listwise Rerankers with Content-Agnostic Probability Calibration — CapCal is a training-free content-agnostic calibration that decouples positional bias from ranking decisions in generative listwise rerankers. DOI arXiv
ICML Ranking Free RAG: Replacing Re-ranking with Selection in RAG for Sensitive Domains — METEORA swaps relevance re-ranking for rationale-driven evidence selection with an elbow-detection cutoff and a verifier LLM, more accurate with fewer chunks. arXiv
ICML Rethinking the Reranker: Boundary-Aware Evidence Selection for Robust Retrieval-Augmented Generation — BAR-RAG reframes the RAG reranker as a boundary-aware evidence selector targeting the generator's usable zone, ~10% end-to-end QA gain under noisy retrieval. arXiv
ICML Very Efficient Listwise Multimodal Reranking for Long Documents — ZipRerank shortens input via query-image early interaction and scores all candidates in one forward pass, up to 10x lower reranking latency on MMDocIR. arXiv
ICTIR Adaptive Re-Ranking — Adjusts re-ranking depth/effort per query rather than a fixed cutoff. DOI
ICTIR Does Form Affect Function? An Extended Study of LLM Re-Ranking Behavior — Probes whether reasoning rerankers are stable across prompt/format. DOI
ICTIR Entity Labels Are Not Entity Signals: A Framework for Observable Relevance in Document Re-Ranking — Separates surface entity labels from signals that drive relevance. DOI
ICTIR Rank, Don't Generate: Statement-level Ranking for Explainable Recommendation — Ranks candidate explanation statements instead of generating text. DOI
ICTIR Rank-ICL: Ranking-based In-context Learning for Search Result Explanation — Ranking-based in-context examples to explain results. DOI
ICTIR Ranking Passages in Relevant Documents Using LLMs — Intra-document passage ranking from LLM relevance signals. DOI
SIGIR Consensus-Anchored Expansion for Noise-Resilient Re-ranking — Anchors on consensus signals to keep re-ranking robust to noise. DOI
SIGIR CRED: Calibrated Relational Enhanced Distillation for LLM-Based Pointwise Reranking — Distillation to make pointwise LLM reranking calibrated and cheaper. DOI
SIGIR Domain-Specific Reranking: When is Adaptation Worth the Cost? — Empirically maps when domain adaptation of rerankers pays off. DOI
SIGIR From Unstructured to Structured: LLM-Guided Attribute Graphs for Entity Search and Ranking — Builds attribute graphs with LLMs to rank entities. DOI
SIGIR Learning to Summarize for Search Relevance with Reinforcement Learning — RL-tuned summaries used as a search-relevance signal. DOI
SIGIR On the Robustness of LLM Re-Rankings — Stress-tests how stable LLM re-ranking is under perturbation. DOI
SIGIR One Pass, Any Order: Position-Invariant Listwise Reranking for LLM-Based Recommendation — Removes position bias from listwise LLM reranking for rec. DOI
SIGIR Rank-R1: Enhancing Reasoning in LLM-based Document Rerankers via Reinforcement Learning — RL-trained reranker that reasons over query and candidates before ordering. DOI arXiv
SIGIR rerankers: A Lightweight Unified Toolkit for Reranking Approaches — Open toolkit unifying many reranking methods behind one API. DOI
SIGIR Revisiting Text Ranking in Deep Research — Re-examines classic text ranking inside deep-research pipelines. DOI
SIGIR Temporal User-Agnostic Ranking: Detecting Preference Evolution while Preserving Ethical Principles — Ranks over time without per-user profiling. DOI
SIGIR The Vulnerability of LLM Rankers to Prompt Injection Attacks — Shows LLM rankers can be gamed by injected instructions in documents. DOI
SIGIR Think When Needed: Model-Aware Reasoning Routing for LLM-based Ranking — Spends reasoning only on queries that need it; routes the rest to cheap scoring. DOI
SIGIR When Search Is Not Enough: Ranking Content for Query-Less Browse Surfaces — Ranking where there is no query, only a browse surface. DOI
SIGIR When Vision Meets Texts in Listwise Reranking — Listwise LLM reranking that jointly reasons over visual and textual candidates. DOI

2 · Dense / sparse / multi-vector retrieval & learnt representations

ACL CLEAR: Cross-Lingual Enhancement in Retrieval via Reverse-training — A reverse-training loss using English passages as a bridge to improve cross-lingual dense retrieval, up to 15% gains for low-resource languages. DOI arXiv
ACL ReasonEmbed: Enhanced Text Embeddings for Reasoning-Intensive Document Retrieval — ReMixer data synthesis plus Redapter self-adaptive learning; ReasonEmbed-Qwen3-8B sets a record nDCG@10 of 38.1 on BRIGHT. arXiv
ACL Situated Embedding Models for Context-Aware Dense Retrieval — SitEmb represents short chunks conditioned on a broader context window to improve dense retrieval over long documents. arXiv
ACL Findings Hybrid-Vector Retrieval for Visually Rich Documents: Combining Single-Vector Efficiency and Multi-Vector Accuracy — HEAVEN's two-stage single-then-multi-vector framework keeps 99.87% of multi-vector Recall@1 while cutting per-query compute by 99.82%. arXiv
ICML AOEB: Benchmarking Agent-Oriented Multimodal Embeddings — A multi-task multimodal embedding benchmark for agent-centric retrieval (code, tool, reasoning, memory) beyond MTEB. ICML
ICML HypRAG: Hyperbolic Dense Retrieval for Retrieval Augmented Generation — Hyperbolic dense retrievers in Lorentz space with geometry-aware pooling, beating Euclidean baselines on MTEB and RAGBench. ICML arXiv
ICML ModernVBERT: Towards Smaller Visual Document Retrievers — A 250M-parameter vision-language encoder for visual document retrieval that matches models up to 10x larger, CPU-efficient. ICML arXiv
ICTIR A Replicability Study of XTR — Reproduces and stress-tests the XTR multi-vector retriever. DOI
ICTIR Analysis of Reasoning-Intensive Retrieval in Japanese — Reasoning-heavy retrieval beyond English. DOI
ICTIR Effectiveness of Sparse Retrieval for Automated Fact-Checking — Revisits learned-sparse retrieval for evidence finding. DOI
ICTIR Model Merging as an Alternative to Fine-Tuning for a Multi-Domain Dense Retriever — Merges domain experts instead of fine-tuning on combined data. DOI
SIGIR A Parametric Memory Head for Continual Generative Retrieval — Memory head for continually-updating generative retrieval. DOI
SIGIR A Replicability Study of Joint Product Quantisation for Space-Efficient Dense Retrieval — Replicates joint PQ for compact dense retrieval. DOI
SIGIR A Voronoi Cell Formulation for Principled Token Pruning in Late-Interaction Retrieval Models — Geometry-grounded token pruning for ColBERT-style retrieval. DOI
SIGIR ARHN: Answer-Centric Relabeling of Hard Negatives with Open-Source LLMs for Dense Retrieval — LLM-relabeled hard negatives for dense retrieval. DOI
SIGIR Auditing Query Drift: Do Users Actually Benefit from Pseudo-Relevance Feedback? — Questions whether PRF actually helps users. DOI
SIGIR Benchmarking Filtered Approximate Nearest Neighbor Search on Transformer-based Embedding Vectors — Benchmark of filtered ANN over transformer vectors. DOI
SIGIR Beyond Hard Negatives: The Importance of Score Distribution in Knowledge Distillation — Score distribution, not just negatives, matters in distillation. DOI
SIGIR Clustering-Based Methods for Vector-Based Pseudo-Relevance Feedback — Cluster-based PRF for dense retrieval. DOI
SIGIR Comparing Token Pruning Approaches for Multi-Vector Retrieval — Head-to-head study of late-interaction token pruning. DOI
SIGIR Constructing Hard-Positive Query-Document Pairs for Dense Retrieval via Phrase Representativeness — Builds hard positives to train dense retrievers. DOI
SIGIR Efficient Continual Adaptation of Generative Retrieval Systems over Dynamic Corpora — Keeps generative retrieval current as corpora change. DOI
SIGIR Efficient Sparse Retrieval with Lightweight Superblock Pruning — Superblock pruning to speed up learned-sparse retrieval. DOI
SIGIR From Tokens to Concepts: Leveraging SAE for SPLADE — Sparse autoencoders to interpret and improve SPLADE. DOI
SIGIR Insights into the Efficiency of Open-Source Score-at-a-Time Search Engines (Reproducibility) — Reproducibility study of score-at-a-time engines. DOI
SIGIR Internalizing Explicit Reasoning into Latent Space for Dense Retrieval — Self-distills chain-of-thought into the dense embedding. DOI
SIGIR LACONIC: Dense-Level Effectiveness for Scalable Sparse Retrieval via Two-Phase Training — Two-phase curriculum lifts sparse retrieval to dense quality. DOI
SIGIR Multi-Vector Index Compression in Any Modality — Modality-agnostic compression for multi-vector indexes. DOI
SIGIR Negative Data Mining for Contrastive Learning in Dense Retrieval at IKEA.com — Industrial negative mining for dense retrieval. DOI
SIGIR PrepRet: Automated Data Preparation Pipeline Selection for Neural Retrieval — Auto-selects data-prep pipelines for neural retrieval. DOI
SIGIR RankEvolve: Automating the Discovery of Retrieval Algorithms via LLM-Driven Evolution — LLM-driven evolutionary search for retrieval algorithms. DOI
SIGIR Reproduction Beyond Benchmarks: ConstBERT and ColBERT-v2 Across Backends and Query Distributions — Reproduces late-interaction models across backends. DOI
SIGIR Revisiting BM25 Feedback Models using HyDE — Combines classic PRF with hypothetical-document embeddings. DOI
SIGIR Scalable K-Means Guided Partitioning for Block-based Sparse Document Retrieval — Cluster-guided partitioning for faster sparse retrieval. DOI
SIGIR Scaling Laws for Embedding Dimension in Information Retrieval — First principled scaling curve for retrieval embedding size. DOI arXiv
SIGIR Semantic Recall for Vector Search — Improving recall in vector search. DOI
SIGIR Sparton: Fast and Memory-Efficient Triton Kernel for Learned Sparse Retrieval — GPU kernel making learned-sparse retrieval fast and lean. DOI
SIGIR The Matryoshka Hypencoder — Nested-dimension hypernetwork encoder for flexible embeddings. DOI
SIGIR Why Advanced Encoders Lag on Sparse Retrieval? Bridging Vocabulary Gaps — Diagnoses and closes the vocabulary gap for strong encoders. DOI

3 · Agentic retrieval, deep research & RAG

ACL AgentGL: Towards Agentic Graph Learning with LLMs via Reinforcement Learning — Equips LLM agents with graph-native tools and search-constrained thinking for multi-scale exploration via RL. DOI
ACL CheckRLM: Effective Knowledge–Thought Coherence Checking in Retrieval-Augmented Reasoning — Extracts factual claims from reasoning chains to detect and correct knowledge inconsistencies at inference. DOI
ACL GTA: Generating Long-horizon Tasks for Web Agents at Scale — Formalizes multi-hop web-agent task generation and releases a dynamic, reproducible benchmark. DOI
ACL Rethinking Reasoning-Intensive Retrieval: Evaluating and Advancing Retrievers in Agentic Search Systems — Introduces the BRIGHT-Pro benchmark and RTriever-Synth corpus, LoRA-tuning RTriever-4B for static and agentic search protocols. DOI arXiv
ACL Findings Agentic Conversational Search with Contextualized Reasoning via Reinforcement Learning — A conversational agent interleaving search and reasoning across turns, RL-trained for evolving user goals over four benchmarks. DOI arXiv
ACL Findings Agentic-R: Learning to Retrieve for Agentic Search — Trains a retriever with local query-passage relevance plus global answer correctness, beating baselines on seven single- and multi-hop QA benchmarks. DOI
ACL Findings AgenticRAGTracer: A Hop-Aware Benchmark for Diagnosing Multi-Step Retrieval Reasoning in Agentic RAG — A hop-aware, LLM-constructed benchmark for step-by-step diagnosis of multi-step retrieval reasoning in agentic RAG. DOI
ACL Findings PRA-RAG: Provably Robust Aggregation in Retrieval-Augmented Generation against Retrieval Corruption — Samples retrieved-text combinations and exploits embedding geometry to derive a robust aggregate with theoretical bounds against RAG poisoning. DOI arXiv
ICML DECOR: Learning to Decompose and Collaborate in Deep Search via Multi-Agent Reinforcement Learning — Multi-agent RL that decomposes deep-search tasks and coordinates collaborating agents on complex information-seeking queries. ICML
ICML DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research — An open long-form deep-research agent trained with rubrics that co-evolve with the policy to give discriminative on-policy feedback. ICML arXiv
ICML From Conflict to Consensus: Boosting Medical Reasoning via Multi-Round Agentic RAG — MA-RAG turns semantic conflict among candidate answers into new retrieval queries across agentic rounds for medical QA. ICML arXiv
ICML Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning — Builds a lightweight knowledge hypergraph and models retrieval as multi-turn agent-environment interaction, optimised end-to-end with RL. ICML arXiv
ICTIR As We May Search — Vision piece re-imagining search in the agentic era. DOI
ICTIR Beyond Relevance: On the Relationship Between Retrieval and RAG Information Coverage — Retrieval relevance and answer coverage are different objectives. DOI
ICTIR DeepResearchGym: A Free, Transparent, and Reproducible Sandbox for Deep Research — Open, reproducible evaluation environment for deep-research agents. DOI
ICTIR Plan-and-Refine in RAG: Diverse and Comprehensive Responses via Global Exploration and Local Exploitation — Global+local search for more complete RAG answers. DOI
ICTIR Preventing Content Leakage in LLM-Based Medical RAG: Structure-Only Retrieval — Structure-only retrieval for leak-free clinical summarization. DOI
ICTIR RAAD: Retrieval-Augmented Ambiguity Detection via Answer Diversity — Uses answer diversity to flag ambiguous questions. DOI
ICTIR RoTRAG: Rule of Thumb Reasoning for Conversation Harm Detection with RAG — Retrieves heuristics to detect harmful conversation. DOI
ICTIR Towards Adaptive and Retriever-friendly Retrieval-augmented Generation via Reinforcement Learning — RL-tunes generation to be easier for the retriever to support. DOI
ICTIR Uncertainty Quantification for Multimodal Retrieval Augmented Generation — Confidence estimates across retrieval and generation in multimodal RAG. DOI
SIGIR Agentic Search in the Wild: Intents and Trajectory Dynamics from 14M+ Real Search Requests — Large-scale empirical characterization of how agents search. DOI arXiv
SIGIR AgentSim: A Platform for Verifiable Agent-Trace Simulation — Simulation platform for reproducible agent traces. DOI
SIGIR CroSearch-R1: Better Leveraging Cross-lingual Knowledge for Retrieval-Augmented Generation — Cross-lingual knowledge use in RAG. DOI
SIGIR Deep Search with Hierarchical Meta-Cognitive Monitoring Inspired by Cognitive Neuroscience — Adds explicit hierarchical metacognition to deep-search control. DOI
SIGIR Evaluation of Agents under Simulated AI Marketplace Dynamics — Evaluates agents inside a simulated AI marketplace. DOI
SIGIR Exploration-and-Thinking: Agentic Reasoning over Knowledge Graphs via LLM-RL Synergy — LLM+RL agent that reasons over knowledge graphs. DOI
SIGIR Fast and Faithful: Real-Time Verification for Long-Document RAG Systems — Real-time faithfulness verification for long-document RAG. DOI
SIGIR Hierarchical Embedding Fusion for Retrieval-Augmented Code Generation — Fuses embeddings for retrieval-augmented code generation. DOI
SIGIR HiRA: Decoupling Planning and Execution with Hierarchical Reasoning in Deep Search — Separates strategic planning from specialized execution in deep search. DOI
SIGIR How Should Multimodal Information Be Chunked for Complex Question Answering? — Studies chunking strategies for multimodal QA. DOI
SIGIR Incentivizing Retrieval-Augmented Generation via Inner Adaptive Context Selection — Learns to select context inside the RAG loop. DOI
SIGIR Learning from Natural Language Feedback for Personalized Question Answering — Uses NL feedback instead of scalar rewards for personalized QA. DOI
SIGIR Learning to Retrieve from Agent Trajectories — Trains the retriever directly on agent interaction data. DOI arXiv
SIGIR Learning to Trust: Dynamic Utilization of RAG for E-commerce Search Relevance — Decides when to trust RAG for e-commerce relevance. DOI
SIGIR MCP Servers for Pyserini and RankLLM: Enabling Agentic Retrieval-Augmented Generation — MCP server wrappers to plug IR tools into agents. DOI
SIGIR Mixture-of-Retrieval Experts for Reasoning-Guided Multimodal Knowledge Exploitation — Routes among specialized retrievers for multimodal reasoning. DOI
SIGIR Numerical Hallucinations in Retrieval-Augmented Generation: Detection and Analysis — Detects and analyzes number errors in RAG output. DOI
SIGIR Personalized Deep Research: A User-Centric Framework, Dataset, and Hybrid Evaluation — Personalizes deep research with a dataset and hybrid evaluation. DOI
SIGIR Policy-Guided RAG: Enforcing Verbatim and Controlled Synthesis — Constrains RAG generation to policy-verbatim synthesis. DOI
SIGIR PurifAI: Detecting and Fixing Search-Induced Distortions in Web-Augmented LLMs — Cleans distortions introduced by web search into LLM answers. DOI
SIGIR Question-Adaptive Graph Learning for Multi-hop Retrieval Augmented Generation — Graph learning tailored per question for multi-hop RAG. DOI
SIGIR Reward Shaping for Robust Refusal in Small Language Models for RAG QA — Teaches small models to refuse when RAG evidence is weak. DOI
SIGIR SmartSearch: Process Reward-Guided Query Refinement for Search Agents — Process rewards and dual-level credit to improve agent queries. DOI
SIGIR SRAG: A Lightweight and Specialized Retrieval-augmented Generation System at the Edge — Compact RAG system that runs at the edge. DOI
SIGIR STRIDE: Strategic Iterative Decision-Making for Retrieval-Augmented Multi-Hop QA — Iterative decision-making for multi-hop RAG question answering. DOI
SIGIR Towards Knowledgeable Deep Research: Framework and Benchmark — New framework and benchmark for knowledge-grounded deep research. DOI
SIGIR WebMall - A Multi-Shop Benchmark for Evaluating Web Agents — Benchmark for shopping web agents. DOI

4 · Query & intent understanding, search behavior

SIGIR A Comparative Study of Users' Information-Seeking Across Search Engines and Generative AI Chatbots — Search vs. chatbot information-seeking behavior. DOI
SIGIR Agentic Query Reformulation for Contextualized Hyper-Personalized Product Search — Agentic reformulation for product search. DOI
SIGIR An Eye Tracking Study: Are AI Overviews Changing Search Behavior? — Eye-tracking evidence on AI Overviews' effect on attention. DOI
SIGIR Automating Generation of Long-Form Queries — Synthesizes human-style long queries. DOI
SIGIR Behavioral Feature Boosting via Substitute Relationships for E-commerce Search — Substitute-aware features for e-commerce search. DOI
SIGIR Beyond Semantic Similarity: Explicit Intent Modeling for Query-Product Matching — Explicit intent for query-product matching. DOI
SIGIR Characterising the Effects of Social Conformity on Information Seeking in Mixed Human-AI Environments — Social conformity in human-AI search. DOI
SIGIR Cognitive Style Shapes Search Behaviours: An fNIRS Study of Exploratory Search — Brain-imaging study of exploratory search. DOI
SIGIR Cross-Sensory Comparison of EEG Signals for Brain-Based Information Retrieval — EEG signals for brain-based IR. DOI
SIGIR Demand-Calibrated Facet Diversification with Deployable Soft Quotas for E-Commerce Search — Facet diversification with soft quotas. DOI
SIGIR Evaluating Proactivity in Conversational Search through Anticipatory Retrieval — Measures proactive/anticipatory conversational search. DOI
SIGIR Following the Eye-Tracking Evidence: Established Web-Search Assumptions Fail in Carousel Interfaces — Position-bias assumptions break in carousels. DOI
SIGIR From Lab to Reality: An Eye-tracking Study of How Users are Influenced to Search in the Era of GenAI — How GenAI influences search behavior, in situ. DOI
SIGIR How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews — Measures GenAI's effect on real search outcomes. DOI
SIGIR IPQA: A Benchmark for Core Intent Identification in Personalized Question Answering — Benchmark for personalized-QA intent. DOI
SIGIR KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao — Multimodal personalized search in production. DOI
SIGIR NeurPIU: Neurobiologically Inspired Personalized Intent Understanding in LLMs — Bio-inspired personalized intent modeling. DOI
SIGIR Policy-Grounded Dynamic Facet Suggestions for Job Search — Dynamic facet suggestions for job search. DOI
SIGIR Recasting Web-Scale Query Suggestion as Dense Retrieval — Query suggestion via dense retrieval. DOI
SIGIR SkyDistill: Navigating Fuzzy Flight Search Ranking via Precise Intent Distillation — Intent distillation for flight-search ranking. DOI
SIGIR Synthetic Data Powers Product Retrieval for Long-tail Knowledge-Intensive Queries — Synthetic data for long-tail product retrieval. DOI
SIGIR Understanding and Modeling Heterogeneous Search Behavior — Models the diversity of real search strategies. DOI
SIGIR User Search Behavior and Knowledge Effects in a Hybrid Search Environment — Knowledge effects in hybrid (classic+AI) search. DOI
SIGIR What Gets Cited: Competitive GEO in AI Answer Engines — Generative-engine optimization in AI answer engines. DOI
SIGIR When & How to Write for Personalized Demand-aware Query Rewriting in Video Search — Personalized query rewriting for video search. DOI

5 · Evaluation, relevance judgments & LLM-as-judge

ACL HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing — A large Chinese writing benchmark over 12 genres showing LLM-as-judge practices are vulnerable to textual disturbance. DOI
ACL LiveCultureBench: a Multi-Agent, Multi-Cultural Benchmark for LLMs in Dynamic Social Simulations — Embeds LLMs as agents in a simulated town, scoring task completion and adherence to socio-cultural norms. DOI
ACL Findings Recovered in Translation: Efficient Pipeline for Automated Translation of Benchmarks and Datasets — An efficient pipeline for automated benchmark/dataset translation, validated with reference-based and LLM-as-judge evaluation. DOI
ICML A Judge-Aware Ranking Framework for Evaluating LLMs without Ground Truth — Improves human-preference agreement in ground-truth-free evaluation by modelling judge reliability and calibrating ranking uncertainty. ICML
ICML QEDBench: Quantifying the Alignment Gap in Automated Evaluation of University-Level Mathematical Proofs — First benchmark contrasting course rubrics with expert criteria for grading undergraduate math proofs. ICML
ICML Quantifying Biases in LLM-as-a-Judge Evaluations — Systematically measures biases in LLM-as-a-judge autograders used as a scalable stand-in for human evaluation. ICML
ICML Who can we trust? LLM-as-a-jury for Comparative Assessment — A Bradley-Terry extension with per-judge discriminator parameters to jointly infer rankings and judge reliability. ICML
ICTIR ARIC: A Cognitive Framework for Explanatory Narrative Evaluation in CIS — Cognitive framework for evaluating explanations in conversational IR. DOI
ICTIR Bridging the Gap between Subsampled and Full-Corpus Evaluation — Reconciles cheap subsampled with full-corpus evaluation. DOI
ICTIR DoGMaTiQ: Automated Generation of Question-and-Answer Nuggets for Report Evaluation — Auto-generates nuggets to evaluate long-form reports. DOI
ICTIR LLM-Driven Usefulness Judgment for Web Search Evaluation — Moves from relevance to usefulness as the LLM-judged signal. DOI
ICTIR Proposing an Evaluation Framework for Legal Case Retrieval Beyond Binary Relevance — Bibliographic-coupling approach to graded legal-retrieval evaluation. DOI
ICTIR Reasoning with Large Language Models for Relevance Judgements — Tests reasoning-augmented LLMs as relevance assessors. DOI
ICTIR Search Arena Meets Nuggets: Explanations and Diagnostics for LLM Response Evaluation — Arena comparison plus nugget diagnostics. DOI
ICTIR To Believe or Not To Believe: Supporting-Information Tools for Judging AI Veracity — Which supporting tools help humans judge AI truthfulness. DOI
SIGIR Decision-Theoretic Stopping Rules for Document Screening — When to stop screening under a decision-theoretic objective. DOI
SIGIR Formalized Information Needs Improve Large-Language-Model Relevance Judgments — Formalizing the need sharpens LLM label quality. DOI
SIGIR G-CoS: An Interpretable Gain-Cost Framework for User Satisfaction Estimation in Generative IR — Interpretable gain-cost model of user satisfaction. DOI
SIGIR How Variability Influences Podcast Search: Queries, Transcriptions, and Judges — Transcription and judge variability move podcast-search results. DOI
SIGIR Humans, LLMs, and Measures Do Not Align in Attributed Information Retrieval — Human, LLM, and metric judgments diverge on attributed IR. DOI
SIGIR Hybrid Pooling with LLMs via Relevance Context Learning — LLM-assisted pooling to build test collections cheaply. DOI
SIGIR Learning to Rank with Multi-Criteria LLM-Judge Annotations — Trains rankers from multi-criteria LLM judgments. DOI
SIGIR Mine over Yours: How Authorship Biases Evaluation in Generative Information Retrieval — Shows authorship bias in generative-IR evaluation. DOI
SIGIR Stop Using the Wilcoxon Test: Myth, Misconception and Misuse in IR Research — Methodological critique of a staple significance test. DOI
SIGIR The Powerless Noise: How Experimental Settings Shape the Reported Power of Noise — How experimental design distorts reported statistical power. DOI
SIGIR Topic-Specific Classifiers are Better Relevance Judges than Prompted LLMs — Lightweight classifiers can out-judge prompted LLMs. DOI
SIGIR Towards a Relevance Posterior in Neural Information Access — Argues for a posterior over relevance, not point labels. DOI
SIGIR When LLM Judges Inflate Scores: Exploring Overrating in Relevance Assessment — Documents systematic over-rating by LLM judges. DOI

6 · Fact-checking, verification & factuality

ACL LLM-Generated Text May Harm Your Retrieval! A Robust Detection Strategy for Retrieval-Augmented Generation — A robust detection strategy against LLM-generated text contaminating the external corpora RAG relies on. DOI
ACL When Efficiency Meets Safety: A Benchmark Security Analysis of KV Cache Compression in LLMs — Benchmarks the security implications of KV-cache compression methods in LLMs. DOI
ACL Findings Streaming Hallucination Detection in Long Chain-of-Thought Reasoning — Treats hallucination in long CoT as an evolving latent state, enabling real-time streaming detection with evidence. DOI arXiv
SIGIR AdversarialCoT: Single-Document Retrieval Poisoning for LLM Reasoning — Shows retrieval-poisoning attacks on LLM reasoning. DOI
SIGIR Attend to Fragments: How Key Information Affects LLMs for Factual Inconsistency Detection — Fragment attention for inconsistency detection. DOI
SIGIR Deja Vu in Plots: Cross-Session Evidence with Retrieval-Augmented LLMs for Live Streaming Risk Assessment — Cross-session RAG for live-streaming risk. DOI
SIGIR EVADE-Bench: Multimodal Benchmark for Evaluating and Enhancing Evasive Content Detection — Benchmark for evasive-content detection. DOI
SIGIR ExDR: Explanation-driven Dynamic Retrieval Enhancement for Multimodal Fake News Detection — Explanation-driven retrieval for multimodal fake news. DOI
SIGIR JARVIS: An Evidence-Grounded Retrieval System for Interpretable Deceptive Reviews Adjudication — Evidence-grounded detection of deceptive reviews. DOI
SIGIR Mitigating Adversarial Attacks by Transferring LLM-generated Narrative Reasoning for Robust Fake News Detection — Narrative reasoning transfer for fake-news robustness. DOI
SIGIR Multi-Sourced, Multi-Agent Evidence Retrieval for Fact-Checking — Multi-agent evidence retrieval for fact-checking. DOI
SIGIR Retrieval-Augmented Multimodal Model for Fake News Detection — RAG-style multimodal fake-news detection. DOI
SIGIR Towards Evidence-Aware Retrieval and Verification for Scientific Fact-Checking — Evidence-aware retrieval for science claims. DOI

7 · Self-improving & self-evolving systems

ACL Aligning Large Language Models via Fully Self-Synthetic Data — SAO generates and self-evaluates all prompts, responses and preferences for alignment with no external data. DOI arXiv
ACL EVOTOOL: Self-Evolving Tool-Use Policy Optimization via Blame-Aware Mutation and Diversity-Aware Selection — Optimizes a modular tool-use policy through a self-evolving loop of trajectory collection, blame attribution and targeted mutation. DOI
ACL Improving Long-Context Translation via Self-Supervised Dual Learning — LongDu self-improves long-document translation via back-translation reconstructability rewards, annotation-free. DOI
ACL Demo AgentFactory: A Self-Evolving Framework Through Executable Subagent Accumulation and Reuse — A demo framework whose library of executable subagents grows and improves over time to cut effort on similar tasks. DOI
ACL Findings A Self-Evolving LLM Agent Framework for Role-Based Norm Compliance in Healthcare — A healthcare agent with structured memory and dual-layer reflection that updates its prompts via preference-style optimization with AI feedback. DOI
ACL Findings Self-Evolving Multi-Agent Systems via Textual Backpropagation — Agentic Neural Network: multi-agent teams self-evolve roles, prompts and coordination via a backprop-like backward phase. DOI arXiv
ACL Findings Towards Self-Improving Error Diagnosis in Multi-Agent Systems — ErrorProbe attributes semantic failures in multi-agent systems and updates memory only after successful verification. DOI arXiv
ACL Findings TT-SI: Self-Improving LLM Agents with Test-Time Training — Adapts an agent on the fly by test-time training on self-augmented uncertain instances: +5.48% accuracy with 68x fewer training samples. DOI arXiv
ICML From Interactions to Principles: Experience-Driven Self-Distillation for Evolving LLM Agents — EvolveR distills an agent's own interaction history into reusable strategic principles and feeds them back via self-distillation. ICML
ICML Large Language Model Agents Are Not Always Faithful Self-Evolvers — Shows self-evolving agents that reuse past experience can fail to faithfully reflect it — a caution on trusting self-reported improvement. ICML
ICML On the Generalization Gap in Self-Evolving Language Model Reasoning — Measures how far self-generated supervision sits from oracle supervision across four self-evolution methods on controlled logical reasoning. ICML
ICML SE-GA: Memory-Augmented Self-Evolution for GUI Agents — A GUI agent that pairs hierarchical memory with an iterative self-improvement pipeline (test-time memory extension plus memory-augmented self-evolution training). ICML

8 · Recursive self-improvement & self-play

ACL OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment — Generates synthetic rubrics at scale for reward modeling and alignment — a self-reward-style pipeline. DOI
ACL Findings Improving LLM Code Reasoning via Semantic Equivalence Self-Play with Formal Verification — A generator (Alice) and evaluator (Bob) train adversarially, refereed by formal proofs of program equivalence. DOI arXiv
ICML Anchoring Self-Play for Code Repair — Generator-fixer self-play in a single model, anchored to a reference set so injected bugs stay close to real-world ones. ICML
ICML Evolving Quantitative Reasoning through Self-Play in Digital Twin Markets — Agents test and adapt strategies via self-play in a controllable digital-twin market, forming a bounded self-evolving loop. ICML
ICML MemEvolve: Meta-Evolution of Agent Memory Systems — Jointly evolves an agent's experiential knowledge and its memory architecture, so the system improves how it learns, not just what it knows. ICML
ICML Toward Training Superintelligent Software Agents through Self-Play SWE-RL — One LLM trained by RL self-play to inject and repair increasingly hard bugs specified by test-suite improvements; beats a human-data baseline on SWE-bench. ICML arXiv

9 · Self-reward, self-verification & self-consistency

ICML Conversation for Non-verifiable Learning: Self-Evolving LLMs through Meta-Evaluation — CoNL unifies generation, evaluation and meta-evaluation via multi-agent self-play, beating self-rewarding baselines without external judges. ICML
ICML Optimal Self-Consistency for Efficient Reasoning with Large Language Models — Derives power-law scaling for self-consistency and analyzes fixed- versus dynamic-allocation sampling for chain-of-thought. ICML
ICML SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning — Turns self-supervised objectives into dense, automatic, verifiable rewards for RL fine-tuning — no human preferences or AI judges. ICML
ICML When to Trust the Cheap Check: Weak and Strong Verification for Reasoning — Analyzes when weak verification (self-consistency, proxy rewards) suffices versus when a strong verifier is required. ICML

10 · Self-correction, self-reflection & test-time RL

ACL Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling — REFORM searches for reward-inconsistent variants and fine-tunes the reward model on a targeted augmentation of its own failures. DOI
ACL Findings Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection — ORION refines teacher chains-of-thought via error-aware self-reflection that incorporates the student's own mistakes. DOI arXiv
ACL Findings Invocation Refiner: A Plug-and-Play Module for Rectifying LLM Tool Invocations — Takes a raw model output plus prompt and query and synthesizes corrected tool invocations. DOI
ACL Findings LLMs as ASP Programmers: Self-Correction Enables Task-Agnostic Nonmonotonic Reasoning — An automated loop where structured feedback from an ASP solver drives iterative refinement of LLM-generated programs. DOI
ACL Findings QueueEDIT: Structural Self-Correction for Sequential Model Editing in LLMs — A structural mapping loss plus a queue that realigns previously edited parameters to self-correct degradation in sequential editing. DOI arXiv
ICML Beyond Majority Voting: Self-Reflective Test-Time Reinforcement Learning for LLM Reasoning — SR-TTRL uses self-reflective verification to produce high-fidelity pseudo-labels for test-time RL instead of majority voting. ICML
ICML Learning to Correct: Reinforcement Learning for Multi-Attempt Chain-of-Thought — Trains reasoning models to make multiple verified attempts with self-reflection, improving self-correction over long chains. ICML
ICML Test-Time Reinforcement Learning for Flow Matching — Flow-TTRL aligns text-to-image generation on the fly by reinterpreting intermediate latents as an implicit policy. ICML

11 · Self-training & self-distillation

ICML Self-Distillation Enables Continual Learning — Self-Distillation Fine-Tuning learns on-policy directly from demonstrations, supporting continual learning without catastrophic forgetting. ICML
ICML Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models — OPSD has one LLM act as both teacher and student under different contexts to improve reasoning. ICML

12 · Reasoning & test-time compute

ACL Less Languages, Less Tokens: An Efficient Unified Logic Cross-lingual Chain-of-Thought Reasoning Framework — UL-XCoT cuts >50% decoding cost across 18 languages while keeping competitive cross-lingual reasoning. DOI arXiv
ACL Look Light, Think Heavy: What Multimodal Chain-of-Thought Reasoning Can and Cannot Do — Finds multimodal CoT handles verbal reflection but lacks deep visual introspection. DOI
ACL Findings Decoupling the Effect of Chain-of-Thought Reasoning: A Human Label Variation Perspective — Reassesses the measured effect of chain-of-thought through the lens of human label variation. DOI
ICML AsyncSpade: Efficient Test-Time Scaling with Asynchronous Sparse Decoding — Cuts the KV-cache bottleneck of long chain-of-thought test-time scaling with asynchronous sparse decoding, >20% faster per token at matched accuracy. ICML
ICML HERMES: Towards Efficient and Verifiable Mathematical Reasoning in LLMs — A tool-assisted agent interleaving informal reasoning with formally verified Lean proofs, raising accuracy while spending fewer tokens. ICML
ICML UniScale: Adaptive Unified Inference Scaling via Online Joint Optimization of Model Routing and Test-Time Scaling — Casts model routing plus test-time scaling as a contextual bandit over the quality-cost trade-off. ICML

13 · Agents, tool use & multi-agent systems

ICML AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Conditions — Systematically evaluates tool-using agents under both user-side instruction noise and tool-side result noise. ICML
ICML AppWorld-UL: Benchmarking Diverse Agent-User Interactions for Tool-Use — 306 user-in-the-loop tasks needing diverse agent-user interaction across nine simulated apps. ICML
ICML ExCyTIn-Bench: Evaluating LLM Agents on Cyber Threat Investigation — Evaluates agents on cyber threat investigation using questions derived from investigation graphs. ICML
ICML Multi²: Hierarchical Multi-Agent Decision-Making with LLM-Based Agents in Interactive Environments — Decomposes agent behavior into complementary hierarchical roles for decision-making in interactive environments. ICML
ICML SciAgentGym: Benchmarking Multi-Step Scientific Tool-Use in LLM Agents — A scalable environment with 1,780 domain tools across four natural sciences plus a tiered agentic tool-use benchmark. ICML

14 · RL, RLHF & preference optimization

ACL Reward Alignment Optimization: A Direct Point-wise Alignment Approach — RAO uses an explicit reward model and a prefix-consistency principle for point-wise direct alignment. DOI
ACL Findings AgentV-RL: Scaling Reward Modeling with Agentic Verifier — Trains a verifier that interleaves tool use with internal reasoning via RL to scale reward modeling. DOI
ACL Findings Curriculum-RLAIF: Curriculum Alignment with Reinforcement Learning from AI Feedback — Presents AI-feedback preference data in order of increasing difficulty during alignment. DOI
ICML Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation — Diagnoses and mitigates advantage-collapse failure modes in GRPO-style RL fine-tuning. ICML
ICML Autoregressive Direct Preference Optimization — ADPO folds autoregressive structure explicitly into the DPO objective. ICML
ICML Distributionally Robust Reinforcement Learning with Human Feedback — Distributionally robust formulations for both reward-based RLHF and reward-free DPO. ICML
ICML Multi-Objective Preference Optimization: Improving Human Alignment of Generative Models — MOPO maximizes a primary preference objective under tunable lower bounds on secondary objectives via KL-regularized constraints. ICML
ICML Reliability-Aware LLM Alignment from Inconsistent Human Feedback — Tackles DPO's tendency to overfit ambiguous, high-disagreement preference pairs by modelling annotator reliability. ICML

15 · Alignment, safety & interpretability

ICML Jailbreak Foundry: From Papers to Runnable Attacks for Reproducible Benchmarking — A multi-agent system that turns published jailbreak papers into executable attack modules for reproducible safety benchmarks. ICML
ICML LALM-as-a-Judge: Benchmarking Large Audio-Language Models for Safety Evaluation in Multi-Turn Spoken Dialogues — Benchmarks large audio-language models as safety judges in multi-turn spoken dialogue. ICML
ICML MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety — 10,389 adversarial prompts spanning 2,665 distinct harmful intents for multi-turn jailbreak evaluation. ICML
ICML Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation — A bi-level adversarial training scheme for zero-shot jailbreak defense, mostly <5% attack success across three LLMs. ICML

16 · Efficiency: distillation, MoE, long-context & decoding

ACL HybridKV: Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference — Hybrid KV-cache compression to speed up multimodal LLM inference. DOI
ACL LongSpec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification — Lossless speculative decoding for long-context inference with efficient drafting and verification. DOI
ICML Mean Flow Distillation: Robust and Stable Distillation for Flow Matching Models — A stable distillation method for flow-matching models, SoTA on 4D occupancy forecasting and text-to-image. ICML
ICML ReMoE: Boosting Expert Reuse through Router Fine-Tuning in Memory-Constrained MoE Inference — Router fine-tuning to raise token-wise expert reuse for fine-grained MoE under memory-constrained inference. ICML
ICML SoftMoE: Soft Differentiable Routing for Mixture-of-Experts in LLMs — Replaces discrete MoE routing with a truncated soft top-k relaxation, enabling gradient-based routing while activating fewer experts. ICML
ICML Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention — Identifies critical KV entries as a byproduct of verification and loads only those when drafting subsequent tokens. ICML

17 · Multilinguality & low-resource

ACL LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance — Language-adaptive hints let RL improve multilingual reasoning without harming language consistency. DOI
ACL VFA: Empowering Multilingual MLLMs via Vision-Free Adaptation — Decouples multilingual language enhancement from visual alignment by composing complementary task vectors over a shared backbone. DOI

18 · Theory: optimization & generalization

ICML Dynamics and representation structure of local approximations to gradient-based learning in linear recurrent neural networks — Uses dynamical-systems theory on data-aligned linear RNNs to compare stability and convergence of local gradient approximations. ICML
ICML Generalization and Scaling Laws for Mixture-of-Experts Transformers — A generalization/scaling theory for MoE Transformers that separates active per-input capacity from routing combinatorics. ICML
ICML Understanding Generalization and Forgetting in In-Context Continual Learning — Extends single-task in-context-learning theory to task sequences to characterize generalization and forgetting. ICML

19 · Perspectives & position papers

SIGIR Emancipatory Information Retrieval: Towards Critical IR Theories and Practices — Critical-theory agenda for IR. DOI
SIGIR From Search to Ask to Act: The Evolution of Information Access in the Age of Large Models and Agents — Perspective on the search-to-agent evolution. DOI
SIGIR Multi-Agentic Recommender Systems: Foundations, Perspectives, and Lessons from Large-Scale Deployments — Perspective on multi-agent recommenders. DOI
SIGIR Question Answering Fit for Purpose: A Perspective From NLP and User Modeling — Perspective on purpose-fit QA. DOI

20 · Learning-to-rank theory, bias, click models & fairness  ·  less central

ICTIR A Theoretical Framework for Risk Analysis of Stochastic Rankers — Formal risk analysis for rankers that sample orderings. DOI
ICTIR Exposure-Based Reinforcement Learning to Rank — RL-to-rank driven by exposure rather than raw clicks. DOI
ICTIR From Noise to Order: Learning to Rank via Denoising Diffusion — Casts ranking as a denoising-diffusion process. DOI
ICTIR Prior-Data Fitted Networks as Tabular Foundation Models for Ranking in Low-Data Settings — Brings PFNs to ranking where labels are scarce. DOI
ICTIR Relaxed Term Matching for Neural Axiomatic Diagnostics — Axiomatic diagnostics adapted to neural rankers. DOI
ICTIR Stability in Competitive Search with Results Diversification — Game-theoretic stability when content competes. DOI
SIGIR Adaptive Autoguidance for Item-Side Fairness in Diffusion Recommender Systems — Item-side fairness for diffusion recommenders. DOI
SIGIR Additive Control Variates Dominate Self-Normalisation in Off-Policy Evaluation — Better estimators for off-policy IR evaluation. DOI
SIGIR An Epistemic Position-Based Click Model: From Interactions to Distributions of Relevance and Bias — Click model yielding distributions over relevance and bias. DOI
SIGIR CLAX: Fast and Flexible Neural Click Models in JAX — Fast neural click-model library. DOI
SIGIR Diagnosing Identifiability in Two-Tower Models for Unbiased Learning to Rank — Identifiability conditions for two-tower unbiased LTR. DOI
SIGIR LearnDCG: End-to-End Joint Optimization of Ranker and Loss in Neural Ranking — Jointly learns the ranker and its ranking loss. DOI
SIGIR Logit Inflation in ListMLE: Theoretical Analysis and Mitigation Strategies — Diagnoses and fixes a pathology in the ListMLE loss. DOI
SIGIR OPS: An Order-Preserving Sorting Network for Information Retrieval — Differentiable order-preserving sorting for ranking. DOI
SIGIR Post-hoc Provider Fairness Adaptation via Hierarchical Exposure Alignment — Exposure-aligned provider fairness applied post-hoc. DOI
SIGIR Unifying On- and Off-Policy Variance Reduction Methods — Unified variance reduction for counterfactual/LTR estimation. DOI
SIGIR Variance Reduction for Heavy-Tailed Monetization Metrics in Ranking Experiments — Post-stratification for noisy ranking A/B metrics. DOI

21 · Multimodal, cross-modal & structured retrieval  ·  less central

ICTIR Adaptive Token-Aware Query Reformulation for Text-to-Image Retrieval — Token-aware reformulation for cross-modal retrieval. DOI
SIGIR A Sketch+Text Composed Image Retrieval Dataset for Thangka — Composed image retrieval dataset (Thangka art). DOI
SIGIR Agentic Spatio-Temporal Grounding via Collaborative Reasoning — Agentic grounding across space and time. DOI
SIGIR Attention Grounded Enhancement for Visual Document Retrieval — Grounds attention to improve document-image retrieval. DOI
SIGIR Beyond Unimodal Perspectives: Generative Retrieval with Multimodal Semantics — Generative retrieval with multimodal semantics. DOI
SIGIR CoMCo: Consistency-Aware Multi-Agent Coordination for Zero-Shot Cross-Modal Entity Matching — Multi-agent zero-shot cross-modal matching. DOI
SIGIR Corpus-Centric Learning for Zero-Shot Table Retrieval — Zero-shot retrieval over tabular corpora. DOI
SIGIR FAST-MEL: Fast, Accurate, Storage-Efficient Multimodal Entity Linking — Efficient multimodal entity linking. DOI
SIGIR FollowTable: A Benchmark for Instruction-Following Table Retrieval — Benchmark for instruction-conditioned table retrieval. DOI
SIGIR From Interference to Stability: Video Moment Retrieval with Relevance Feedback — Relevance feedback for video moment retrieval. DOI
SIGIR Generation-Augmented Video Corpus Moment Retrieval — Reframes moment retrieval as generative understanding. DOI
SIGIR Learning Noise-Resilient and Transferable Graph-Text Alignment via Dynamic Quality Assessment — Robust graph-text alignment. DOI
SIGIR ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment — Reasoning-guided alignment for visual-doc retrieval. DOI
SIGIR Seeing the Whole Through the Parts: Object Discovery via Semantic Part Mining — Weakly-supervised object discovery. DOI
SIGIR Spectral-Adaptive Adversarial Hashing for Robust Image Retrieval — Adversarially robust hashing for image retrieval. DOI
SIGIR Unifying Granularity and Reliability: Text-based Person Retrieval — Robust, efficient text-based person retrieval. DOI
SIGIR Visual-RAG: Benchmarking Text-to-Image Retrieval Augmented Generation — Benchmark for text-to-image RAG. DOI

22 · Domain & specialized retrieval, resources  ·  less central

ICTIR A Comprehensive Taxonomy of Temporal Dimensions in Natural Language Queries — Organizes how time appears in queries. DOI
ICTIR A Design Science Framework for Human-Agent Teams in Information Retrieval — Design-science view of human+agent IR. DOI
ICTIR Interpretable Legal Similarity: From Embeddings to Obligations — Maps legal embedding similarity to obligations. DOI
ICTIR MedQueryIntent: A Large-Scale Benchmark Dataset for Medical Search Intent Classification — Benchmark for medical query intent. DOI
ICTIR Patent Representation Learning via Self-supervision — Self-supervised patent representations. DOI
ICTIR TopicTune: A Topic Alignment Approach to Improve LLM Reasoning on Social Questions — Topic alignment to steer LLM reasoning. DOI
SIGIR Building Foundations for IR in Extremely Low-Resource Ethnic Languages: Tanzania — IR foundations for Tanzanian languages. DOI
SIGIR CSyMR: Benchmarking Compositional Music Information Retrieval in Symbolic Music Reasoning — Symbolic-music IR benchmark. DOI
SIGIR CTCL: A Cross-Language Benchmark for Matching Patients to Clinical Trials — Patient-to-trial matching benchmark. DOI
SIGIR Enhancing Judgment Document Generation via Agentic Legal Information Collection and Rubric-Guided Optimization — Agentic legal-document generation. DOI
SIGIR Improving Amharic Information Retrieval with Translative and Multi-Agent Debate RAG — Multilingual RAG for Amharic. DOI
SIGIR Is It Novel and Why? Fine-Grained Patent Novelty Prediction Based on Passage Retrieval — Passage retrieval for patent-novelty judgments. DOI
SIGIR KCC: Korean Civil Case Dataset for Legal Information Retrieval — Korean legal-IR dataset. DOI
SIGIR Morphology-Aware Retrieval for Low-Resource Environments: Shona — Morphology-aware retrieval for Shona. DOI
SIGIR Multilingual and Domain-Agnostic Tip-of-the-Tongue Query Generation for Simulated Evaluation — Tip-of-the-tongue query generation. DOI
SIGIR Retrieve, Rerank, Answer, Experiment: Hands-On IR Research with PyTerrier — Hands-on IR research toolkit resource. DOI
SIGIR RULER: Robust Unified LLM-based Efficient Retrieval for Legal Information — Unified LLM retrieval tuned for legal search. DOI
SIGIR SQuTR: A Robustness Benchmark for Spoken Query to Text Retrieval under Acoustic Noise — Spoken-query retrieval robustness benchmark. DOI
SIGIR T-RADAR: Simulating Trademark Examination as an Interactive Retrieval Interface — Trademark-examination retrieval interface. DOI
SIGIR The Wikidata Query Logs Dataset — Large Wikidata query-log resource. DOI
SIGIR ViCSR: A Benchmark and Two-Stage Framework for Vietnamese Case-to-Statute Retrieval — Vietnamese case-to-statute retrieval. DOI

23 · Generative recommendation, semantic IDs & LLM recommenders  ·  less central

SIGIR Adaptive Mix Preference Optimization for Generative Recommendation — Mixed preference optimization for generative rec. DOI
SIGIR AIPO: Adaptive Anchored Intent-aware Policy Optimization for Generative Recommendation — Intent-aware policy optimization for generative rec. DOI
SIGIR BEAR: Towards Beam-Search-Aware Optimization for Recommendation with LLMs — Aligns training with beam search in LLM rec. DOI
SIGIR Beyond Item IDs: Scaling Short-Form-Video Recommendation via Semantic-Native Long Sequence Modeling — Semantic-native sequences for short-video rec. DOI
SIGIR Beyond Static Best-of-N: Bayesian List-wise Alignment for LLM-based Recommendation — Bayesian listwise alignment for LLM rec. DOI
SIGIR Beyond the Single Path: Divergent Reasoning for LLM-based Recommendation — Counters reasoning-path collapse in LLM recommenders. DOI
SIGIR Full Retraining, Incremental Fine-tuning, and Hybrid Serving for Industrial Generative Recommenders — Serving strategies for industrial generative rec. DOI
SIGIR GenRec: A Preference-Oriented Generative Framework for Large-Scale Recommendation — Preference-oriented generative rec at scale. DOI
SIGIR GFlowGR: Fine-tuning Generative Recommendation Frameworks with Generative Flow Networks — GFlowNets for generative recommendation. DOI
SIGIR GNN-Based Item Indexing for LLM-Enhanced Recommendation — GNN item indexing to feed LLM recommenders. DOI
SIGIR ItemRAG: Item-Based Retrieval-Augmented Generation for LLM-Based Recommendation — Item-based RAG for LLM recommenders. DOI
SIGIR Large-Scale Online Learning for Generative List Recommendation in E-commerce — Online learning for generative list rec. DOI
SIGIR Learning Decomposed Contextual Token Representations for Generative Recommendation — Decomposed token representations for generative rec. DOI
SIGIR LWGR: Lagrangian-Constrained Personalized World Knowledge for Generative Recommendation — Injects constrained world knowledge into generative rec. DOI
SIGIR M2GR: Generative User Interest Modeling via Multi-Granularity Multi-Objective CoT — Multi-granularity CoT for industrial generative rec. DOI
SIGIR Mitigating Collaborative Semantic ID Staleness in Generative Retrieval — Keeps semantic IDs fresh in generative retrieval. DOI
SIGIR Modular Representation Compression: Adapting LLM Representations for Efficient Recommendation — Controls modularity of LLM representations for rec. DOI
SIGIR Semantic IDs for Recommender Systems at Snapchat: Use Cases, Challenges, Design Choices — Production use of semantic IDs at Snapchat. DOI
SIGIR SID-Coord: Coordinating Semantic IDs for ID-based Ranking in Short-Video Search — Coordinates semantic IDs with ID-based ranking. DOI
SIGIR Think, But Don't Tell: Implicit Reasoning for Sequential Recommendation via Multi-Teacher Distillation — Distills large-model reasoning into lightweight recommenders. DOI
SIGIR Unifying Search and Recommendation in LLMs via Gradient Multi-Subspace Tuning — One LLM for both search and recommendation. DOI
SIGIR Universal Item Tokenization for Transferable Generative Recommendation — Transferable item tokenization across domains. DOI
SIGIR Verifiable Reasoning for LLM-based Generative Recommendation — Adds verifiable reasoning traces to LLM recommenders. DOI

24 · Recommendation: sequential, multimodal, graph, CF, CTR & ads  ·  less central

ICTIR Equity by Design? On the Trade-Offs in Fairness-Driven Recommendation in Heterogeneous Two-Sided Markets — Analyzes fairness trade-offs across two-sided markets. DOI
ICTIR RAMP: Robust Ad Recommendation Under Limited Personalized-Feature Availability via Masking and Alignment — Masking + alignment for ad rec when personalization is sparse. DOI
ICTIR RecQuest: Towards Estimating User Domain Knowledge in Conversational Recommender Systems — Infers a user's expertise to adapt conversational recommendation. DOI
ICTIR RIDRec: Retrieval-Enhanced Intent Diffusion for Anonymous Short-Session Recommendation — Diffusion over retrieved intents for cold, anonymous sessions. DOI
SIGIR Adaptive Rich-kernelized Contrastive Learning for Collaborative Filtering — Kernelized contrastive CF. DOI
SIGIR AsarRec: Adaptive Sequential Augmentation for Robust Self-supervised Sequential Recommendation — Augmentation for robust self-supervised seq rec. DOI
SIGIR Beyond Dense Connectivity: Explicit Sparsity for Scalable Recommendation — Explicit sparsity for scalable rec. DOI
SIGIR Bridging Behavior and Semantics for Time-aware Cross-Domain Sequential Recommendation — Time-aware cross-domain sequential rec. DOI
SIGIR Contrastive Flow Matching for Collaborative Filtering — Flow matching applied to collaborative filtering. DOI
SIGIR Countering Interest Over-Smoothing: Distilling Latent Factors via Diffusion for Multi-Interest Retrieval — Diffusion to preserve diverse user interests. DOI
SIGIR Debiased Recommendation Beyond the Positive Propensity Assumption — Debiasing without the positive-propensity assumption. DOI
SIGIR DIGEST: Dynamic Graph Refinement with Dual Contrastive Semantic Transfer for Multimodal Recommendation — Dynamic graph refinement for multimodal rec. DOI
SIGIR DIVER: Unlocking Diversity in Ad Headline Generation with Large Language Models — Diverse ad-headline generation. DOI
SIGIR Equip Pre-ranking with Target Attention by Residual Quantization — Target attention for efficient pre-ranking. DOI
SIGIR Exploring Test-time Scaling via Prediction Merging on Large-Scale Recommendation — Test-time scaling by merging predictions. DOI
SIGIR From Clues to Generation: Language-Guided Conditional Diffusion for Cross-Domain Recommendation — Language-guided diffusion for cross-domain rec. DOI
SIGIR From Raw Features to Effective Embeddings: Multimodal Recipe Recommendation — Three-stage embeddings for recipe rec. DOI
SIGIR Fusion and Alignment Enhancement with LLMs for Tail-item Sequential Recommendation — LLM help for long-tail items. DOI
SIGIR Generative Auto-Bidding with Unified Modeling and Exploration — Generative auto-bidding for ads. DOI
SIGIR HyFormer: Revisiting Sequence Modeling and Feature Interaction in CTR Prediction — Rethinks CTR sequence + feature interaction. DOI
SIGIR Inferring Targets from Calibrated Hesitations in Multi-Behavior Recommendation — Uses hesitation signals in multi-behavior rec. DOI
SIGIR MLLMRec: A Preference Reasoning Paradigm with Graph Refinement for Multimodal Recommendation — Preference reasoning + graph refinement for multimodal rec. DOI
SIGIR Optimizing Marketing Subsidies via Counterfactual Learning with Asymmetric Reward — Counterfactual learning for subsidy allocation. DOI
SIGIR Revisiting Collaborative Filtering by Unleashing the Power of Similarity — Rethinks similarity in collaborative filtering. DOI
SIGIR Robust Multimodal Recommendation via Graph Retrieval-Enhanced Modality Completion — Completes missing modalities via graph retrieval. DOI
SIGIR SpecTran: Spectral-Aware Transformer-based Adapter for LLM-Enhanced Sequential Recommendation — Spectral adapter for LLM-enhanced sequential rec. DOI
SIGIR Structural and Disentangled Adaptation of Large VLMs for Multimodal Recommendation — Adapts VLMs for multimodal rec. DOI
SIGIR Understanding DNNs in Feature Interaction Models: A Dimensional Collapse Perspective — Explains feature-interaction models via dimensional collapse. DOI

25 · Diffusion & flow-based generation  ·  less central

ICML A Kinetic-Energy Perspective of Flow Matching — Introduces Kinetic Path Energy, an action-like diagnostic of accumulated effort along a flow ODE trajectory. ICML
ICML High-Accuracy Sampling for Diffusion Models and Log-Concave Distributions — Outstanding Paper: training-free first-order rejection sampling giving polylog(1/δ)-step high-accuracy diffusion sampling from scores alone. ICML arXiv
ICML The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models — Outstanding Paper questioning whether arbitrary generation order actually helps diffusion language models. ICML

26 · Multimodal & speech  ·  less central

ACL "I See What You Did There": Can Large Vision-Language Models Understand Multimodal Puns? — Probes whether large vision-language models can understand multimodal puns. DOI
ACL ReLook: Vision-Grounded RL with a Multimodal LLM Critic — A vision-aware multimodal LLM critic supplies scores and feedback inside a vision-grounded RL loop. DOI

My reading, by research topic

What the cross-venue view makes obvious

1 Self-improvement is venue-agnostic

2 Retrieval learned to reason — everywhere

3 The judge and the verifier are the shared bottleneck

Every catalogued paper, grouped by topic

1 · Reasoning-first ranking & LLM rerankers

2 · Dense / sparse / multi-vector retrieval & learnt representations

3 · Agentic retrieval, deep research & RAG

4 · Query & intent understanding, search behavior

5 · Evaluation, relevance judgments & LLM-as-judge

6 · Fact-checking, verification & factuality

7 · Self-improving & self-evolving systems

8 · Recursive self-improvement & self-play

9 · Self-reward, self-verification & self-consistency

10 · Self-correction, self-reflection & test-time RL

11 · Self-training & self-distillation

12 · Reasoning & test-time compute

13 · Agents, tool use & multi-agent systems

14 · RL, RLHF & preference optimization

15 · Alignment, safety & interpretability

16 · Efficiency: distillation, MoE, long-context & decoding

17 · Multilinguality & low-resource

18 · Theory: optimization & generalization

19 · Perspectives & position papers

20 · Learning-to-rank theory, bias, click models & fairness  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

21 · Multimodal, cross-modal & structured retrieval  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

22 · Domain & specialized retrieval, resources  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

23 · Generative recommendation, semantic IDs & LLM recommenders  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

24 · Recommendation: sequential, multimodal, graph, CF, CTR & ads  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

25 · Diffusion & flow-based generation  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

26 · Multimodal & speech  ·  <span style="font-weight:400;color:var(--muted)">less central</span>

Sources & method note

What the cross-venue view makes obvious

1 Self-improvement is venue-agnostic

2 Retrieval learned to reason — everywhere

3 The judge and the verifier are the shared bottleneck

1 · Reasoning-first ranking & LLM rerankers

2 · Dense / sparse / multi-vector retrieval & learnt representations

3 · Agentic retrieval, deep research & RAG

4 · Query & intent understanding, search behavior

5 · Evaluation, relevance judgments & LLM-as-judge

6 · Fact-checking, verification & factuality

7 · Self-improving & self-evolving systems

8 · Recursive self-improvement & self-play

9 · Self-reward, self-verification & self-consistency

10 · Self-correction, self-reflection & test-time RL

11 · Self-training & self-distillation

12 · Reasoning & test-time compute

13 · Agents, tool use & multi-agent systems

14 · RL, RLHF & preference optimization

15 · Alignment, safety & interpretability

16 · Efficiency: distillation, MoE, long-context & decoding

17 · Multilinguality & low-resource

18 · Theory: optimization & generalization

19 · Perspectives & position papers

20 · Learning-to-rank theory, bias, click models & fairness &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

21 · Multimodal, cross-modal & structured retrieval &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

22 · Domain & specialized retrieval, resources &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

23 · Generative recommendation, semantic IDs & LLM recommenders &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

24 · Recommendation: sequential, multimodal, graph, CF, CTR & ads &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

25 · Diffusion & flow-based generation &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

26 · Multimodal & speech &nbsp;·&nbsp; <span style="font-weight:400;color:var(--muted)">less central</span>

Sources & method note

20 · Learning-to-rank theory, bias, click models & fairness · <span style="font-weight:400;color:var(--muted)">less central</span>

21 · Multimodal, cross-modal & structured retrieval · <span style="font-weight:400;color:var(--muted)">less central</span>

22 · Domain & specialized retrieval, resources · <span style="font-weight:400;color:var(--muted)">less central</span>

23 · Generative recommendation, semantic IDs & LLM recommenders · <span style="font-weight:400;color:var(--muted)">less central</span>

24 · Recommendation: sequential, multimodal, graph, CF, CTR & ads · <span style="font-weight:400;color:var(--muted)">less central</span>

25 · Diffusion & flow-based generation · <span style="font-weight:400;color:var(--muted)">less central</span>

26 · Multimodal & speech · <span style="font-weight:400;color:var(--muted)">less central</span>