Contents

 

ForewordxiiiPreface           xv Acknowledgmentsxix Introduction              xxi

 

 

Part I    From Traditional Information Retrieval to Modern RAG      1

 

1                       InformationRetrieval   3

1.1                  Definition and Historical Evolution     3

1.1.1              Definition     3

1.1.2              Historical Evolution     4

1.2                  Information Retrieval Components      14

1.2.1              Overview     14

1.2.2              Pre-retrieval      16

1.2.3              Retrieval Models     21

1.2.4              Post-retrieval     30

1.3                  Information Retrieval Applications      34

1.3.1              Search Systems     35

1.3.2              Recommendation Systems    38

1.3.3              Example: WANDS Dataset      40

1.4                  Challenges with IR Systems     45

1.4.1              Lack of Context Awareness     45

1.4.2              Cold-start Problem     45

1.4.3              Scalability     45

1.4.4              Difficulty in Handling Complex Queries      45

1.5                  Summary     46

References     46

 

2                       Large Language Models     49

2.1                  LLMs Overview     49

2.1.1              Evolution of ML Paradigms      49

2.1.2              LLM Fundamentals       54

2.1.3              LLM Families     75

2.1.4              LLM Paradigms     83

 

viii     Contents
2.2	LLM Use Case in Information Retrieval      91
2.2.1	Use LLM for Search Systems     91
2.2.2	Use LLM for Recommendation Systems     102
2.3	Challenges of LLMs for Information Retrieval      108
2.3.1	Core Limitations: Hallucinations and Knowledge Staleness	108
2.3.2	The RAG Solution: Dynamic Knowledge Integration      110

2.3.3              Practice: LLM for Search and Recommendation Tasks      114

2.3.4              Practice: Contextual LLM Responses      115

2.4            Summary     119

References     119

 

3                       Retrieval-augmented Generation      123

3.1                  RAG Overview     123

3.1.1              RAG Components      123

3.1.2              From Tokens to Passage Embeddings      124

3.1.3              RAG as a Retrieval-agnostic Architecture      125

3.2                  Data Preparation and Indexing     126

3.2.1              Understanding Data Source Types     126

3.2.2              Data Preprocessing     129

3.2.3              Chunking Techniques     131

3.2.4              Indexing     136

3.3                  Retrieval Approaches      139

3.3.1              The Distinctive Nature of RAG Retrieval      140

3.3.2              Core Retrieval Methods in RAG     140

3.3.3              Advanced Retrieval Patterns      145

3.3.4              Retrieval Performance Optimization      146

3.3.5              Reranking and Filtering       147

3.4                  Generation     148

3.4.1              Generation Pipeline      149

3.4.2              Model Selection Framework      150

3.4.3              Parameter Tuning     150

3.4.4              Inference Speed and Optimization Techniques      151

3.5                  Summary     153

References     153

 

4                       Practice: RAG Implementation       157

4.1                  Overview of LangChain and LlamaIndex      157

4.1.1              LangChain and LangGraph     157

4.1.2              LlamaIndex      158

4.1.3              Comparison Summary     159

4.2                  Implementing RAG Pipelines with LlamaIndex      159

4.2.1              Offline Stages: Loading and Indexing      160

4.2.2              Online Stage: Querying      166

4.2.3              Key Takeaways     175

4.3                  Example: Implementing RAG on the WANDS Dataset      175

4.3.1              Indexing Phase     175

4.3.2              Query Phase     176

Contents  ix

 

4.3.3              Key Takeaways     189

4.4                  Implementing Agentic RAG     189

4.4.1              Offline Index Construction      190

4.4.2              Offline Agent Construction      191

4.4.3              Online Query Processing      192

4.4.4              Key Takeaways     195

4.5                  Summary     195

References     196

 

Part II Graphs and Knowledge Graphs     197

 

5                       GraphsandGraphDatabases     199

5.1                  Introduction to Graphs     199

5.1.1              Types of Graphs     200

5.1.2              Graph Representation      207

5.1.3              Graph Operations     210

5.1.4              Graph Algorithms      213

5.1.5              Applications of Graphs     216

5.2                  Graph Databases: Comparison and Analysis     217

5.2.1              Notable Graph Databases     218

5.2.2              Comparison Summary     221

5.3                  Introduction to Neo4j and Cypher      221

5.3.1              Cypher Basic Operations      222

5.3.2              Neo4j Utility Functions: APOC     224

5.3.3              Neo4j Graph Analytics: GDS     226

5.4                  Practice: Flight Network Analysis and Optimization      228

5.4.1              Problem Setting      228

5.4.2              Understanding the Network Structure      229

5.4.3              Set Up Database and Load Data     229

5.4.4              Graph Algorithms in Flight Networks      232

5.5                  Summary     244

References     245

 

6                       Knowledge Graphs   247

6.1                  Understanding KGs     247

6.1.1              Core Structure and Components      248

6.1.2              Example: Marvel Universe KG     249

6.1.3              Types of KGs    253

6.1.4              Ontology in KGs     256

6.2                  Construction and Management of KGs     261

6.2.1              Schema Design     261

6.2.2              Data Sources and Extraction Techniques     265

6.3                  KGs Analytics and Enrichment       268

6.3.1              Graph Algorithms in KGs     268

6.3.2              Graph Embeddings     279

6.3.3              Graph Neural Networks      282

6.4                  LLMs and KGs     287

x     Contents

 

6.4.1              Comparisons and Unifications     288

6.4.2              Leveraging LLMs to Enrich KGs     288

6.4.3              Using KGs to Enhance LLMs: A Preview of Graph-RAG     292

6.5                  Practice: Constructing KG from the WANDS Dataset      293

6.5.1              LlamaIndex Neo4j Integration      294

6.5.2              Extract Query-product Triples     294

6.5.3              Extract Product-Class Triples     296

6.5.4              Extract Product-category Triples     298

6.5.5              Load Documents into Neo4j     300

6.5.6              Key Takeaways     300

6.6                  Practice: Construct KG from Unstructured Data      301

6.6.1              Key Takeaways     302

6.7                  Summary     303

References     304

 

Part III Integrate RAG with Graph     307

 

7                       Graph-basedRetrieval-augmentedGeneration      309

7.1                  Introduction to Graph-RAG     309

7.1.1              The Graph-RAG Paradigm      309

7.1.2              Why Graph-RAG?    310

7.2                  Architecture and Components of Graph-RAG     315

7.2.1              Graph-based Indexing     316

7.2.2              Graph-based Retrieval     318

7.2.3              Graph-based Generation      333

7.3                  Applications      337

7.3.1              Downstream Tasks     337

7.3.2              Application Domains     338

7.4                  Summary     340

References     341

 

8                       Practice: Graph-RAG Implementations       345

8.1                  Graph-RAG on WANDS Dataset with LlamaIndex      345

8.1.1              Querying with Neo4j and Chroma     345

8.1.2              Removing Chroma Dependency     351

8.1.3              Key Takeaways     352

8.2                  Graph-RAG on Wiki and Kaggle Data with LangChain      352

8.2.1              Wikipedia Text Search and Question Answering      352

8.2.2              Answering Kaggle CSV Recommendation Questions      357

8.3                  Summary     364

Reference     365

 

9                       Graph-RAG Evaluations      367

9.1                  Performance Metrics Framework      367

9.1.1              Retrieval-specific Metrics     367

9.1.2              Generation Quality Metrics      372

9.1.3              Graph-specific Metrics     376

 

		Contents	xi
9.1.4        System Performance Metrics     380 9.2                  Quality Assessment Methodologies     382 9.2.1              Human Evaluation Framework      383 9.2.2              Automated Evaluation Methods     386 9.2.3              Meta-evaluation      389 9.2.4              Best Practices for Quality Assessment      391 9.3                  Benchmarking Frameworks      392 9.3.1              Established Benchmarks      392 9.3.2              Benchmark Design Principles       394 9.3.3              Comparative Analysis Framework      395 9.4                  Tools and Platforms for Graph-RAG Evaluations      396 9.4.1              Tools and Platforms Overview     396 9.4.2              Integration Best Practices     399 9.4.3              Future Directions in Tooling      400 9.5                  Practice: Evaluating RAG Pipelines Using Ragas     400 9.5.1              Preparing the Evaluation Dataset      400 9.5.2              Defining Evaluation Metrics     401 9.5.3              Executing the Evaluation Process     402 9.5.4              Key Takeaways     402 9.6            Summary      402 References     403
PartIVAdvancedImplementationsandFrontiers      405
10                   Graph-RAGFrameworksforEnhancedInformationRetrieval      407 10.1               Overview of Graph-RAG Frameworks for Search and Recommendations	407
10.1.1           RAPTOR: Recursive Abstractive Processing for Tree-organized Retrieval 10.1.2           G-Retriever: Joint Text-graph Embeddings      411 10.1.3           Microsoft GraphRAG: Global Sensemaking Through Community Summaries     414 10.1.4           HippoRAG: Neurobiologically-inspired Long-term Memory      417 10.2               Graph-RAG Tools and Softwares Overview     421 10.2.1           What Is Available in the Current Market for Graph-RAG?      421 10.2.2           When to Choose Which Framework?     423 10.3               Practice: Run Graph-RAG Frameworks      424 10.3.1           RAPTOR     424 10.3.2           G-Retriever     429 10.3.3           Microsoft GraphRAG     432 10.3.4           HippoRAG     444 10.4         Summary      453 References     454	408
11          Frontiers of Graph-RAG      455

11.1               Emerging Trends in Graph-RAG      455

11.1.1           Multimodal Knowledge Graphs      455

11.1.2           Temporal and Dynamic Graphs     462

11.1.3           Advanced GNNs for RAG     468

 

xii    Contents
11.2	Ethical Considerations and Bias Mitigation      478
11.2.1	The Dual Nature of Bias in Graph-RAG      478
11.2.2	Comprehensive Mitigation Strategies     479
11.2.3	Ethical AI Governance Framework      481
11.3	Future Research Directions      482
11.3.1	Explainable Graph-RAG: Making AI Reasoning Transparent	482
11.3.2	Automated Knowledge Graph Construction and Maintenance	483
11.3.3	RL for Intelligent Graph Traversal     485
11.3.4	Federated Learning for Privacy-preserving Graph-RAG      486
11.3.5	Transformative Industry Applications      488
11.4	Summary      490
11.5	Conclusion and Final Thoughts      491 References     492
AA.1 A.1.1 A.1.2 A.2 A.3 A.4 A.5	Set Up Experiment Servers495 Two Models Hosted by vLLM      497 Chat Model Service     497 Embedding Model Service      498 LLM Gateway    499 UI for LLM Models     500 Vector Store     500 Graph Database     501
B B.1 B.2 B.3 B.4 B.4.1 B.4.2 B.4.3 B.5	PrepareSyntheticRecommendationDatafromWANDS       503 Step 1: Download the WANDS Dataset      503 Step 2: Load the Dataset     503 Step 3: Prepare the Data for Recommendation Tasks    504 Step 4: Generate Synthetic Recommendation Data     504 Session-based Recommendation Data     507 User-query Synthetic Profiles      508 Next-item Prediction Sequences      508 Step 5: Dataset Statistics and Sample Entries      509
	Index  511