Advanced Retrieval-Augmented Generation

Bridging Large Language Models and Knowledge Graphs

Inbunden, Engelska, 2026

1 520 kr

Kommande

Beskrivning

Build Accurate, Grounded, and Trustworthy AI Systems with Retrieval-Augmented Generation Large language models are powerful—but they hallucinate. Advanced Retrieval-Augmented Generation offers a complete guide from the foundations of information retrieval (IR) to the cutting-edge frontiers of RAG. Bridging large language models (LLMs) and knowledge graphs (KGs), this book provides the theoretical principles, practical techniques, and hands-on frameworks needed to build reliable AI systems that minimize hallucinations and improve factual correctness. The book covers core concepts of Graph-RAG with applications across search, recommendation, and enterprise AI. Practical chapters demonstrate implementations using LlamaIndex, Neo4j, and leading Graph-RAG frameworks. Readers will learn: IR and LLM fundamentals — model paradigms, transformer architecture, model families, training techniques, prompt engineering, applications, and limitationsRAG pipeline engineering —chunking, indexing, retrieval, ranking, and generationKG construction and analytics — schema design, extraction techniques, graph algorithms, embeddings, and GNNsGraph-RAG architectures and evaluation — graph-based retrieval, graph-assisted generation, hybrid LLM–KG workflows, frameworks, benchmarks, and metricsEmerging directions — multimodal KGs, dynamic graphs, explainable RAG, RL-based traversal, and enterprise-scale implementationsWith extensive hands-on examples and production-ready patterns, Advanced Retrieval-Augmented Generation is an indispensable resource for AI practitioners, ML engineers, researchers, and architects building the next generation of reliable, knowledge-grounded AI systems.

Produktinformation

Utgivningsdatum:2026-09-15
Format:Inbunden
Språk:Engelska
Antal sidor:540
Förlag:John Wiley & Sons Inc
ISBN:9781394374687

Utforska kategorier

Systemvetenskap och AI inom Data och IT

Mer om författaren

Wendy Ran Wei, PhD, is an expert in AI, ML, and LLMs, specializing in search and recommendation systems. She is a Machine Learning Engineer at Airbnb, where she develops retrieval and ranking models and brings LLM technologies into production. She previously held engineering roles at Meta, Pinterest, and Twitter, building large-scale search and recommendation solutions. Dr. Wei received her PhD in Statistics from The Ohio State University. Huijun Wu, PhD, is an Engineer at Samsung Research America with expertise in large-scale distributed systems and data processing. He received his PhD in Computer Science from Arizona State University.

Innehållsförteckning

Contents ForewordxiiiPreface xv Acknowledgmentsxix Introduction xxi Part I From Traditional Information Retrieval to Modern RAG 1 1 InformationRetrieval 31.1 Definition and Historical Evolution 31.1.1 Definition 31.1.2 Historical Evolution 41.2 Information Retrieval Components 141.2.1 Overview 141.2.2 Pre-retrieval 161.2.3 Retrieval Models 211.2.4 Post-retrieval 301.3 Information Retrieval Applications 341.3.1 Search Systems 351.3.2 Recommendation Systems 381.3.3 Example: WANDS Dataset 401.4 Challenges with IR Systems 451.4.1 Lack of Context Awareness 451.4.2 Cold-start Problem 451.4.3 Scalability 451.4.4 Difficulty in Handling Complex Queries 451.5 Summary 46References 46 2 Large Language Models 49 2.1 LLMs Overview 492.1.1 Evolution of ML Paradigms 492.1.2 LLM Fundamentals 542.1.3 LLM Families 752.1.4 LLM Paradigms 83 viii Contents 2.2LLM Use Case in Information Retrieval 912.2.1Use LLM for Search Systems 912.2.2Use LLM for Recommendation Systems 1022.3Challenges of LLMs for Information Retrieval 1082.3.1Core Limitations: Hallucinations and Knowledge Staleness1082.3.2The RAG Solution: Dynamic Knowledge Integration 110 2.3.3 Practice: LLM for Search and Recommendation Tasks 1142.3.4 Practice: Contextual LLM Responses 1152.4 Summary 119References 119 3 Retrieval-augmented Generation 123 3.1 RAG Overview 1233.1.1 RAG Components 1233.1.2 From Tokens to Passage Embeddings 1243.1.3 RAG as a Retrieval-agnostic Architecture 1253.2 Data Preparation and Indexing 1263.2.1 Understanding Data Source Types 1263.2.2 Data Preprocessing 1293.2.3 Chunking Techniques 1313.2.4 Indexing 1363.3 Retrieval Approaches 1393.3.1 The Distinctive Nature of RAG Retrieval 1403.3.2 Core Retrieval Methods in RAG 1403.3.3 Advanced Retrieval Patterns 1453.3.4 Retrieval Performance Optimization 1463.3.5 Reranking and Filtering 1473.4 Generation 1483.4.1 Generation Pipeline 1493.4.2 Model Selection Framework 1503.4.3 Parameter Tuning 1503.4.4 Inference Speed and Optimization Techniques 1513.5 Summary 153References 153 4 Practice: RAG Implementation 157 4.1 Overview of LangChain and LlamaIndex 1574.1.1 LangChain and LangGraph 1574.1.2 LlamaIndex 1584.1.3 Comparison Summary 1594.2 Implementing RAG Pipelines with LlamaIndex 1594.2.1 Offline Stages: Loading and Indexing 1604.2.2 Online Stage: Querying 1664.2.3 Key Takeaways 1754.3 Example: Implementing RAG on the WANDS Dataset 1754.3.1 Indexing Phase 1754.3.2 Query Phase 176 Contents ix 4.3.3 Key Takeaways 1894.4 Implementing Agentic RAG 1894.4.1 Offline Index Construction 1904.4.2 Offline Agent Construction 1914.4.3 Online Query Processing 1924.4.4 Key Takeaways 1954.5 Summary 195References 196 Part II Graphs and Knowledge Graphs 197 5 GraphsandGraphDatabases 1995.1 Introduction to Graphs 1995.1.1 Types of Graphs 2005.1.2 Graph Representation 2075.1.3 Graph Operations 2105.1.4 Graph Algorithms 2135.1.5 Applications of Graphs 2165.2 Graph Databases: Comparison and Analysis 2175.2.1 Notable Graph Databases 2185.2.2 Comparison Summary 2215.3 Introduction to Neo4j and Cypher 2215.3.1 Cypher Basic Operations 2225.3.2 Neo4j Utility Functions: APOC 2245.3.3 Neo4j Graph Analytics: GDS 2265.4 Practice: Flight Network Analysis and Optimization 2285.4.1 Problem Setting 2285.4.2 Understanding the Network Structure 2295.4.3 Set Up Database and Load Data 2295.4.4 Graph Algorithms in Flight Networks 2325.5 Summary 244References 245 6 Knowledge Graphs 247 6.1 Understanding KGs 2476.1.1 Core Structure and Components 2486.1.2 Example: Marvel Universe KG 2496.1.3 Types of KGs 2536.1.4 Ontology in KGs 2566.2 Construction and Management of KGs 2616.2.1 Schema Design 2616.2.2 Data Sources and Extraction Techniques 2656.3 KGs Analytics and Enrichment 2686.3.1 Graph Algorithms in KGs 2686.3.2 Graph Embeddings 2796.3.3 Graph Neural Networks 2826.4 LLMs and KGs 287 x Contents 6.4.1 Comparisons and Unifications 2886.4.2 Leveraging LLMs to Enrich KGs 2886.4.3 Using KGs to Enhance LLMs: A Preview of Graph-RAG 2926.5 Practice: Constructing KG from the WANDS Dataset 2936.5.1 LlamaIndex Neo4j Integration 2946.5.2 Extract Query-product Triples 2946.5.3 Extract Product-Class Triples 2966.5.4 Extract Product-category Triples 2986.5.5 Load Documents into Neo4j 3006.5.6 Key Takeaways 3006.6 Practice: Construct KG from Unstructured Data 3016.6.1 Key Takeaways 3026.7 Summary 303References 304 Part III Integrate RAG with Graph 307 7 Graph-basedRetrieval-augmentedGeneration 3097.1 Introduction to Graph-RAG 3097.1.1 The Graph-RAG Paradigm 3097.1.2 Why Graph-RAG? 3107.2 Architecture and Components of Graph-RAG 3157.2.1 Graph-based Indexing 3167.2.2 Graph-based Retrieval 3187.2.3 Graph-based Generation 3337.3 Applications 3377.3.1 Downstream Tasks 3377.3.2 Application Domains 3387.4 Summary 340References 341 8 Practice: Graph-RAG Implementations 345 8.1 Graph-RAG on WANDS Dataset with LlamaIndex 3458.1.1 Querying with Neo4j and Chroma 3458.1.2 Removing Chroma Dependency 3518.1.3 Key Takeaways 3528.2 Graph-RAG on Wiki and Kaggle Data with LangChain 3528.2.1 Wikipedia Text Search and Question Answering 3528.2.2 Answering Kaggle CSV Recommendation Questions 3578.3 Summary 364Reference 365 9 Graph-RAG Evaluations 367 9.1 Performance Metrics Framework 3679.1.1 Retrieval-specific Metrics 3679.1.2 Generation Quality Metrics 3729.1.3 Graph-specific Metrics 376 Contentsxi9.1.4 System Performance Metrics 3809.2 Quality Assessment Methodologies 3829.2.1 Human Evaluation Framework 3839.2.2 Automated Evaluation Methods 3869.2.3 Meta-evaluation 3899.2.4 Best Practices for Quality Assessment 3919.3 Benchmarking Frameworks 3929.3.1 Established Benchmarks 3929.3.2 Benchmark Design Principles 3949.3.3 Comparative Analysis Framework 3959.4 Tools and Platforms for Graph-RAG Evaluations 3969.4.1 Tools and Platforms Overview 3969.4.2 Integration Best Practices 3999.4.3 Future Directions in Tooling 4009.5 Practice: Evaluating RAG Pipelines Using Ragas 4009.5.1 Preparing the Evaluation Dataset 4009.5.2 Defining Evaluation Metrics 4019.5.3 Executing the Evaluation Process 4029.5.4 Key Takeaways 4029.6 Summary 402References 403 PartIVAdvancedImplementationsandFrontiers 405 10 Graph-RAGFrameworksforEnhancedInformationRetrieval 40710.1 Overview of Graph-RAG Frameworks for Search and Recommendations 407 10.1.1 RAPTOR: Recursive Abstractive Processing for Tree-organized Retrieval10.1.2 G-Retriever: Joint Text-graph Embeddings 41110.1.3 Microsoft GraphRAG: Global Sensemaking Through Community Summaries 41410.1.4 HippoRAG: Neurobiologically-inspired Long-term Memory 41710.2 Graph-RAG Tools and Softwares Overview 42110.2.1 What Is Available in the Current Market for Graph-RAG? 42110.2.2 When to Choose Which Framework? 42310.3 Practice: Run Graph-RAG Frameworks 42410.3.1 RAPTOR 42410.3.2 G-Retriever 42910.3.3 Microsoft GraphRAG 43210.3.4 HippoRAG 44410.4 Summary 453References 454408 11 Frontiers of Graph-RAG 455 11.1 Emerging Trends in Graph-RAG 45511.1.1 Multimodal Knowledge Graphs 45511.1.2 Temporal and Dynamic Graphs 46211.1.3 Advanced GNNs for RAG 468 xii Contents 11.2Ethical Considerations and Bias Mitigation 47811.2.1The Dual Nature of Bias in Graph-RAG 47811.2.2Comprehensive Mitigation Strategies 47911.2.3Ethical AI Governance Framework 48111.3Future Research Directions 48211.3.1Explainable Graph-RAG: Making AI Reasoning Transparent48211.3.2Automated Knowledge Graph Construction and Maintenance48311.3.3RL for Intelligent Graph Traversal 485 11.3.4Federated Learning for Privacy-preserving Graph-RAG 486 11.3.5Transformative Industry Applications 488 11.4Summary 490 11.5Conclusion and Final Thoughts 491References 492 AA.1 A.1.1A.1.2 A.2A.3A.4A.5Set Up Experiment Servers495 Two Models Hosted by vLLM 497 Chat Model Service 497 Embedding Model Service 498 LLM Gateway 499UI for LLM Models 500Vector Store 500Graph Database 501 BB.1B.2B.3B.4 B.4.1B.4.2B.4.3 B.5PrepareSyntheticRecommendationDatafromWANDS 503Step 1: Download the WANDS Dataset 503Step 2: Load the Dataset 503Step 3: Prepare the Data for Recommendation Tasks 504 Step 4: Generate Synthetic Recommendation Data 504 Session-based Recommendation Data 507User-query Synthetic Profiles 508Next-item Prediction Sequences 508Step 5: Dataset Statistics and Sample Entries 509 Index 511