Graph-Powered Analytics and Machine Learning with TigerGraph [Retail PDF] (Victor Lee, Phuc Kien Nguyen etc.)（Z-Library）

# Graph-Powered Analytics and Machine Learning with TigerGraph ## 【One-Line Pitch】 A practical, hands-on guide for data scientists, engineers, and analysts who want to harness graph databases—specifically TigerGraph—to connect, analyze, and learn from connected data for real business outcomes like customer 360 views, fraud detection, and recommendations. ## 【Book Arc】 - **Opening (~0%–9%)**: Introduces the "connect, analyze, learn" three-stage framework and explains why graph thinking matters for modern analytics. Sets up the book's structure: concept chapters followed by use-case chapters, all built around TigerGraph Cloud and the GSQL query language. - **Early (~9%–25%)**: Covers graph fundamentals—vertices, edges, properties, schemas—and contrasts graphs with relational databases. Key insight: relationships are explicit in graphs, eliminating costly runtime joins. Includes a glossary of essential graph terminology and a first look at GSQL through complete, line-by-line examples. - **Early–Middle (~25%–38%)**: Explores data modeling trade-offs: when to use multiple edges vs. event-as-vertex vs. aggregated edge properties. Discusses schema evolution, entity resolution (similar entities share similar neighborhoods), and introduces "graph power"—the unique capabilities graphs bring to connecting dots across datasets. - **Middle (~38%–53%)**: Dives into the first major use case: Customer 360 + Journey Graphs. Shows how to model customer journeys as sequences of engagements, implement similarity queries in GSQL (Jaccard similarity for finding similar contacts and reported cases), and build real-time 360-degree views of business entities. - **Late (~53%–end)**: Continues with additional use cases—startup investment analysis, fraud and money laundering pattern detection, referrals and recommendations, cybersecurity, and airline route analysis—before moving into graph-powered machine learning methods and entity resolution revisited. ## 【Key Takeaways】 - **Graphs make relationships explicit** (Early): Unlike relational databases where connections require runtime joins and secondary indexes, graph databases store relationships directly, enabling faster traversal and more natural queries for connected data problems. - **Modeling choices carry trade-offs** (Early): Representing interaction events as multiple edges, as event vertices, or as aggregated edge properties each have pros and cons. Event-as-vertex offers maximum flexibility for linking and analytics; aggregated edges reduce complexity but lose granularity. - **Schema evolution is a feature, not a chore** (Early): Adding vertex types, edge types, or properties is straightforward. Converting a property into a vertex type (e.g., country) is a common evolutionary step as business needs grow. - **Entity resolution is naturally graph-friendly** (Early): Similar entities share similar neighborhoods, so similarity algorithms like cosine and Jaccard similarity work directly on graph structure—no elaborate table joins required. - **Customer journeys need sequence-aware modeling** (Middle): CRMs store mixed engagement data in tabular form, making journey sequencing painful. Graphs model sequences directly with edges, enabling both individual journey analysis and aggregate pattern detection. - **GSQL is approachable for SQL programmers** (Middle): The query language uses familiar concepts—SELECT, FROM, WHERE, accumulators—making it accessible. Global accumulators (e.g., @@size_A) track single values; vertex-attached accumulators (e.g., @similarity) store per-vertex results. - **Similarity scoring is practical and tunable** (Middle): Queries like jaccard_nbor_reaction let you specify relationship types, top-k results, and sampling thresholds—giving analysts control over what "similar" means in context. ## 【Reading Tips】 - **Skim the relational database comparisons** (Early): If you already understand why joins are costly, the graph-vs-relational contrast is familiar ground. Focus instead on the graph-specific modeling patterns. - **Deep-read the GSQL examples** (Early–Middle): The line-by-line code walkthroughs are the book's core value. Follow along with TigerGraph Cloud free tier to build muscle memory for the syntax. - **Pay attention to modeling trade-off discussions** (Early): The multi-edge vs. event-vertex vs. aggregated-property decision appears repeatedly in later use cases. Understanding it early pays off. - **Use the use-case chapters as templates** (Middle–Late): Each use case follows a pattern—goal, solution, schema, queries, analytics. Extract the reusable patterns rather than memorizing specifics. - **Skip ahead if you're not a beginner** (Late): The machine learning chapter and entity resolution revisit assume comfort with the earlier material. If you're experienced, jump straight to these. ## 【Coverage Limits】 This guide covers the opening through the middle of the book (approximately 0–53%), including graph fundamentals, data modeling, GSQL basics, and the Customer 360 use case. The later use cases (startup investments, fraud detection, cybersecurity, airline routes) and the machine learning chapters are only partially covered in the available excerpts. ##