NVIDIA Nemotron Nano V2 VL NVIDIA introduces a 12B vision-language model achieving leading OCR performance with strong capabilities in document understanding, long video comprehension, and reasoning tasks. 📝 arxiv.org/abs/2511.03929 👨🏽‍💻 huggingface.co/nvidia/NVIDI...

RAGalyst: Automated Human-Aligned Agentic Evaluation for Domain-Specific RAG Presents an automated framework for evaluating domain-specific RAG systems with human-aligned LLM-as-a-Judge metrics and agentic synthetic QA generation. 📝 arxiv.org/abs/2511.04502

KGFR: A Foundation Retriever for Generalized Knowledge Graph Question Answering Enables zero-shot generalization to unseen KGs through LLM-generated relation descriptions and question-conditioned initialization. 📝 arxiv.org/abs/2511.04093 👨🏽‍💻 github.com/yncui-nju/KGFR

On the Brittleness of CLIP Text Encoders Analyzes how CLIP text encoders react to minor input variations in multimedia retrieval, finding that syntactic and semantic perturbations cause the largest instabilities while brittleness concentrates in trivial surface edits. 📝 arxiv.org/abs/2511.04247

Cache Mechanism for Agent RAG Systems Introduces an annotation-free caching framework that reduces RAG storage to 0.015% of original corpus while achieving 79.8% has-answer rate and 80% latency reduction. 📝 arxiv.org/abs/2511.02919

No-Human in the Loop: Agentic Evaluation at Scale for Recommendation Walmart introduces a multi-agent framework that benchmarks 36 LLMs as judges for complementary-item recommendation without human annotation. 📝 arxiv.org/abs/2511.03051

Generative Sequential Recommendation via Hierarchical Behavior Modeling Presents a generative framework for multi-behavior sequential recommendation with cross-level interaction and sequential augmentation. 📝 arxiv.org/abs/2511.03155 👨🏽‍💻 github.com/wzf2000/GAMER

RAGBOOST: Efficient Retrieval-Augmented Generation with Accuracy-Preserving Context Reuse Proposes an efficient RAG system that employs context ordering, de-duplication, and contextual hints while preserving reasoning accuracy. 📝 arxiv.org/abs/2511.03475 👨🏽‍💻 github.com/Edinburgh-Ag...

CLAX: Fast and Flexible Neural Click Models in JAX Introduces a JAX-based library that replaces EM-based optimization with gradient-based methods for click models, achieving orders of magnitude faster training. 📝 arxiv.org/abs/2511.03620 👨🏽‍💻 github.com/philipphager...

MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learning Proposes an agent workflow that maintains compact memory across interactions with higher accuracy and lower overhead. 📝 arxiv.org/abs/2511.02805 👨🏽‍💻 github.com/icip-cas/Mem...

Let Multimodal Embedders Learn When to Augment Query via Adaptive Query Augmentation Proposes a multimodal embedder that intelligently decides when to augment queries rather than augmenting all queries. 📝 arxiv.org/abs/2511.02358

Beyond Single Embeddings: Capturing Diverse Targets With Multi-Query Retrieval Introduces an autoregressive multi-embedding retriever that generates multiple query vectors to capture diverse target distributions. 📝 arxiv.org/abs/2511.02770 👨🏽‍💻 github.com/timchen0618/...

Efficient and Responsible Adaptation of Large Language Models for Robust Top-k Recommendations Proposes a hybrid task allocation framework that strategically uses LLMs to improve recommendations for weak users. 📝 dl.acm.org/doi/10.1145/... 👨🏽‍💻 github.com/i-kiran/TORS...

Zero-RAG: Towards Retrieval-Augmented Generation with Zero Redundant Knowledge Introduces a method to prune redundant knowledge from RAG corpora by 30% while maintaining performance, using Mastery-Score to identify what LLMs already know. 📝 arxiv.org/abs/2511.00505

Taxonomy-based Negative Sampling In Personalized Semantic Search for E-commerce The Home Depot presents a taxonomy-based hard-negative sampling strategy for semantic retrieval in e-commerce that leverages product hierarchies. 📝 arxiv.org/abs/2511.00694

REAR: Retrieve, Expand and Refine for Effective Multitable Retrieval Presents a three-stage LLM-free framework that separates semantic relevance from structural joinability for multi-table retrieval. 📝 arxiv.org/abs/2511.00805

ColMate: Contrastive Late Interaction and Masked Text for Multimodal Document Retrieval Introduces a multimodal document retrieval model with masked OCR pretraining, self-supervised contrastive learning, and TopKSim late interaction. 📝 arxiv.org/abs/2511.00903 👨🏽‍💻 huggingface.co/ahmed-masry/...

Efficient Test-Time Retrieval Augmented Generation Proposes a training-free framework that combines retrieval augmentation with self-consistency mechanisms to improve LLM performance while maintaining efficiency. 📝 arxiv.org/abs/2511.01059

Contextual Relevance and Adaptive Sampling for LLM-Based Document Reranking Proposes a sampling-based uncertainty-aware reranking algorithm that models document relevance as context-dependent rather than fixed. 📝 arxiv.org/abs/2511.01208

RAGSmith: A Framework for Finding the Optimal Composition of Retrieval-Augmented Generation Methods Across Datasets Treats RAG design as an architecture search, using genetic algorithms to optimize pipeline configurations. 📝 arxiv.org/abs/2511.01386 👨🏽‍💻 github.com/yAquila/RAGS...

LiveSearchBench: An Automatically Constructed Benchmark for Retrieval and Reasoning over Dynamic Knowledge Presents a pipeline for constructing retrieval-dependent benchmarks from recent knowledge updates using Wikidata. 📝 arxiv.org/abs/2511.01409 👨🏽‍💻 livesearchbench.github.io

LiCoMemory: Lightweight and Cognitive Agentic Memory for Efficient Long-Term Reasoning Introduces a lightweight hierarchical graph structure for agent memory that achieves accuracy improvement on long-term dialogue benchmarks. 📝 arxiv.org/abs/2511.01448 👨🏽‍💻 github.com/EverM0re/LiC...

CAT-ID2: Category-Tree Integrated Document Identifier Learning for Generative Retrieval In E-commerce Meituan introduces a method for generative retrieval that integrates hierarchical category information into semantic IDs. 📝 arxiv.org/abs/2511.01461 👨🏽‍💻 github.com/lxbdtt/CAT-ID2

Trove: A Flexible Toolkit for Dense Retrieval Introduces an efficient open-source toolkit that simplifies retrieval experiments with on-the-fly data management, flexible modeling, and easy distributed evaluation. 📝 arxiv.org/abs/2511.01857 👨🏽‍💻 ir-trove.dev

A Semi-supervised Co-training Algorithm for Robust Recommendation Proposes SSCDL, a semi-supervised co-training framework that addresses noisy implicit feedback in recommender systems through confidence estimation and iterative cross-co-training. 📝 dl.acm.org/doi/10.1145/...

A Survey on Generative Recommendation: Data, Model, and Tasks Surveys how generative models like LLMs and diffusion models are transforming recommender systems through enhanced data augmentation, model architectures, and novel task capabilities. 📝 arxiv.org/abs/2510.27157

A Survey on Deep Text Hashing: Efficient Semantic Text Retrieval with Binary Representation Surveys deep text hashing methods for efficient semantic text retrieval using compact representations to accelerate search and reduce storage costs. 📝 arxiv.org/abs/2510.27232 👨🏽‍💻 github.com/hly1998/Deep...

SIGMA: Search-Augmented On-Demand Knowledge Integration for Agentic Mathematical Reasoning Presents a multi-agent framework coordinating factual, logical, computational, and completeness perspectives with hypothetical document enhancement. 📝 arxiv.org/abs/2510.27568

Interact-RAG: Reason and Interact with the Corpus, Beyond Black-Box Retrieval Transforms LLM agents from passive query issuers to active manipulators of retrieval through multi-faceted search, anchored matching, and context shaping primitives. 📝 arxiv.org/abs/2510.27566

MARAG-R1: Beyond Single Retriever via Reinforcement-Learned Multi-Tool Agentic Retrieval Introduces a reinforcement-learned multi-tool RAG framework enabling LLMs to dynamically coordinate semantic search, keyword search, filtering, and aggregation. 📝 arxiv.org/abs/2510.27569

I published Vol. 128 of "Top Information Retrieval Papers of the Week" on Substack. 🔗 recsys.substack.com/p/unifying-r...

InfoFlow: Reinforcing Search Agent via Reward Density Optimization Introduces a dual-agent framework that addresses low reward density in deep search training through sub-goal scaffolding, pathfinding hints, and trajectory refinement. 📝 arxiv.org/abs/2510.26575

CRAG-MM: Multi-modal Multi-turn Comprehensive RAG Benchmark Meta presents a benchmark for multi-modal RAG with 6.5K single-turn and 2K multi-turn conversations, focusing on wearable device use cases with egocentric images. 📝 arxiv.org/abs/2510.26160 👨🏽‍💻 huggingface.co/crag-mm-2025

OneTrans: Unified Feature Interaction and Sequence Modeling with One Transformer in Industrial Recommender ByteDance proposes a unified Transformer backbone that jointly performs user-behavior sequence modeling and feature interaction. 📝 arxiv.org/abs/2510.26104

Towards Global Retrieval Augmented Generation: A Benchmark for Corpus-Level Reasoning Introduces a benchmark for evaluating global RAG capabilities requiring corpus-wide analysis across entire document collections. 📝 arxiv.org/abs/2510.26205 👨🏽‍💻 huggingface.co/datasets/Qii...

Seeing Through the MIRAGE: Evaluating Multimodal Retrieval Augmented Generation Introduces a claim-centric evaluation framework for multimodal RAG that assesses information factuality, coverage, citation support, and attribution. 📝 arxiv.org/abs/2510.24870 👨🏽‍💻 github.com/alexmartin17...

Model-Document Protocol for AI Search Introduces a framework that transforms raw documents into structured, LLM-ready knowledge through agentic reasoning, memory grounding, and structured leveraging pathways. 📝 arxiv.org/abs/2510.25160 👨🏽‍💻 github.com/qhjqhj00/MDP...

GReF: A Unified Generative Framework for Efficient Reranking via Ordered Multi-token Prediction Kuaishou proposes an end-to-end generative reranking framework with ordered multi-token prediction, enabling efficient autoregressive generation. 📝 arxiv.org/abs/2510.25220

TV-Rec: Time-Variant Convolutional Filter for Sequential Recommendation Introduces time-variant convolutional filters inspired by graph signal processing to capture position-dependent temporal variations in user sequences. 📝 arxiv.org/abs/2510.25259 👨🏽‍💻 github.com/yehjin-shin/...

Revisiting scalable sequential recommendation with Multi-Embedding Approach and Mixture-of-Experts Kuaishou decomposes item embeddings into multiple lower-dimensional representations and uses mixture-of-experts for adaptive feature transformation. 📝 arxiv.org/abs/2510.25285

Generalized Pseudo-Relevance Feedback Introduces a framework that relaxes relevance and model assumptions in query rewriting by using LLMs with RAG and RL for robust pseudo-relevance feedback. 📝 arxiv.org/abs/2510.25488

MMQ-v2: Align, Denoise, and Amplify: Adaptive Behavior Mining for Semantic IDs Learning in Recommendation Alibaba adaptively aligns behavior and content information while using mixture-of-quantization to generate robust semantic IDs for recommenders. 📝 arxiv.org/abs/2510.25622

UNGER: Generative Recommendation with A Unified Code via Semantic and Collaborative Integration Introduces a generative recommendation method that integrates semantic and collaborative knowledge into a single unified code. 📝 dl.acm.org/doi/10.1145/...

Amazon Nova Multimodal Embeddings: Technical Report and Model Card Amazon presents Nova Multimodal Embeddings, a unified embedding model supporting text, documents, images, video, and audio in a single representation space for cross-modal retrieval. 📝 www.amazon.science/publications...

Repurposing Synthetic Data for Fine-grained Search Agent Supervision Alibaba introduces E-GRPO, a training framework that uses entity information from synthetic data to improve search agent learning through dense rewards based on entity matching. 📝 arxiv.org/abs/2510.24694

Mitigating Hallucination in Large Language Models (LLMs): An Application-Oriented Survey on RAG, Reasoning, and Agentic Systems Introduces a capability-enhancement perspective on hallucination mitigation. 📝 arxiv.org/abs/2510.24476

Graph-Guided Concept Selection for Efficient Retrieval-Augmented Generation Huawei combines core chunk selection with an LLM-independent concept graph to reduce GraphRAG construction costs by 80% while improving retrieval effectiveness. 📝 arxiv.org/abs/2510.24120

DUET: Dual Model Co-Training for Entire Space CTR Prediction Kuaishou introduces a set-wise pre-ranking framework combining dual model co-training with candidate-level interaction modeling to address sample selection bias and improve CTR prediction. 📝 arxiv.org/abs/2510.24369

MiniOneRec: An Open-Source Framework for Scaling Generative Recommendation Presents a fully open-source generative recommendation framework with an end-to-end workflow spanning SID construction and supervised fine-tuning. 📝 arxiv.org/abs/2510.24431 👨🏽‍💻 github.com/AkaliKong/Mi...

Optimizing Retrieval for RAG via Reinforced Contrastive Learning Presents a retrieval framework optimized for RAG through trial-and-feedback reinforced contrastive learning, enabling retrievers to dynamically explore relevance within RAG environments. 📝 arxiv.org/abs/2510.24652