Tag
This paper introduces comprehensive taxonomies for analyzing failures in retrieval-based factuality evaluation in medical AI, revealing fundamental limitations of the retrieve-then-verify paradigm that persist despite model scaling and fine-tuning.
The article proposes HiCoMER, a framework for hierarchical collaborative memory management in LLM agents that improves validity-aware retrieval, reducing outdated memories and enhancing question-answering quality.
Proposes VoxPolyMem, an interaction-aware multimodal memory framework with adaptive agentic retrieval for multi-party spoken conversations, achieving state-of-the-art performance on new benchmarks.
The paper introduces a bias-variance theory for dense retrieval, comparing shared and dual projections, and proposes the CARS method to select optimal geometry based on directional signal estimation from training data.
The paper proposes a claim-gated auditing specification for generative search to address omissions in source relationships, ensuring complete evidence and disclosure for reliable answer interpretation.
CRISS is a retrieval-augmented AI chatbot developed to assist cancer registrars in interpreting complex coding and staging guidelines, evaluated for accuracy and utility in training scenarios.
This paper presents a bilingual evaluation suite for cross-lingual legal question answering in Vietnamese labour law, assessing retrieval, translation, and verifier-guided correction to enhance answer faithfulness and address unsupported claims.
LatWeave introduces a deterministic multi-hop question answering system using knowledge lattices to eliminate hallucination and ensure auditable evidence chains through meet, compare, and abstain operators.
LEGO is a dual-module framework that synergizes Expert GraphRAG and Expert Chain-of-Thought to enhance complex legal reasoning in large language models, achieving superior performance on benchmarks like LawExamQA_Civil.
This paper presents a controlled study on learned context planning for long-context multiple-choice QA, demonstrating that it does not robustly outperform strong retrieval methods like BM25 and hybrid retrieval under various experimental settings.
This paper introduces two taxonomies—decomposing retrieval-stage errors into five quality dimensions and verifier-reasoning errors into six steps—induced automatically via LLM-as-Judge to diagnose where retrieval-based medical factuality evaluation fails. Stress-tests across 4 retrieval methods and 6 frontier verifiers show these failure modes are fundamental limitations of the retrieve-then-verify paradigm, not artifacts of outdated systems.
This paper investigates graph-guided evidence navigation to improve long-novel reasoning with a local 9B language model, providing exploratory evidence from experiments on detective novels without statistically significant gains over baselines.
The paper proposes Wiki Foundation Model (WFM), a novel foundation model for scalable, agent-native representation and retrieval using LLM Wiki, which couples dense text with link structure, achieving strong performance on benchmarks and 10.5x training acceleration.
SCoP introduces a constraint-centric framework for temporal knowledge graph question answering that externalizes temporal decisions to filter evidence and improve answer accuracy on benchmark datasets like MultiTQ and TimelineCronQ-R.
The paper proposes FedRepRAG, a decentralized retrieval-augmented generation framework that exchanges latent representations instead of raw content to reduce computational overhead and improve privacy in sensitive domains like healthcare.
AdaMem introduces a relevance-guided soft-compression framework that adaptively allocates memory tokens in retrieval-augmented generation, outperforming uniform allocation methods with improved accuracy and reduced latency.
The paper introduces Question's Gambit, a first-move retrieval module that improves deep research agent performance by decomposing questions and providing a ranked context before iterative search, showing significant accuracy gains on BrowseComp-Plus.
This study evaluates Beacon, a course-specific Retrieval-Augmented Generation (RAG) system designed to provide module-aligned academic support in higher education, aiming to reduce help-seeking disparities among students by lowering barriers to assistance.
This paper investigates offline multimodal large language models as decision support tools for air operations, detailing a modular retrieval-augmented architecture and presenting a pilot study with the Brazilian Air Force that demonstrates reduced cognitive workload and improved efficiency in doctrinal assessment tasks.
Document Retrieval-Aware Chunking (D-RAC) is a method that normalizes enterprise documents to PDF, converts them to retrieval-optimized Markdown using a multimodal LLM, and chunks them efficiently, significantly reducing token usage and costs compared to agentic chunking.