Graph Theory, Graph Databases, and the Graph Intelligent Platform: Concepts, Development, and Tencent Use Cases

The talk begins by emphasizing that real‑world entities and their relationships can be modeled as graphs, which are essential for representing complex networks like payment and social graphs. It outlines the agenda: thinking like a graph, the youth and simplicity of graph technology, the relationship between graph databases and intelligent platforms, and concrete graph database applications.

1. Thinking Like a Graph – Beyond traditional graph databases, Tencent has developed its own graph computing framework, visualization engine, and analysis components, integrating them with Flink (Oceanus) to support graph‑stream fusion.

2. Graph Theory – Originating with Euler in 1738, graph theory provides the basis for modeling entities (vertices) and relationships (edges). Tencent's EasyGraph, Angel‑Graph, and Graph‑Index platforms implement directed property graphs, enabling storage, computation, visualization, and analysis.

3. Transfer Scenario – In a payment network, entities such as user IDs, bank cards, devices, and IPs become vertices, while transactions and behaviors form edges. Accurate modeling facilitates graph queries, computation, and analytics.

4. The Era of Everything Connected – With the proliferation of social, communication, and payment networks, graph data becomes ubiquitous, making graph‑centric analysis crucial for uncovering hidden patterns.

5. Solving Graph Problems with Graph Techniques – Classical graph algorithms (e.g., adjacency lists, max‑flow) and modern frameworks (Pregel, PowerGraph, GraphX, Gemini) have evolved, leading to the rise of graph neural networks (GNNs) and representation learning.

6. Graph Storage – Graph databases address locality‑aware queries and multi‑hop traversals, supporting both TP and AP workloads, and can be combined with graph computation and visualization for end‑to‑end analysis.

7. Graph Intelligent Platform – Consists of a graph computing framework, database, and visualization engine, integrated with stream processing (Oceanus) to handle billions of vertices and trillions of edges, improving query efficiency for services like WeChat Pay, gaming, and advertising.

8. EasyGraph Architecture – Separates storage and computation, supporting both stateless TP queries and stateful AP analyses, and scales via distributed servers.

9. Graph Visualization – Provides intuitive visual insights for knowledge graphs, financial risk, and network security, employing search, filter, layout, and rendering techniques; large graphs are offloaded to backend for performance.

10. Application Cases – (a) Unified scheduling system: storing task dependencies in a graph database reduced dependency‑checking latency from two minutes to 30 ms, improving real‑time scheduling by 4000×. (b) Financial payment analysis: graph visualization enables rapid local sub‑graph inspection for fund‑flow analysis. (c) Fraud gang detection: a custom PK‑Louvain algorithm with prior knowledge improves detection accuracy by 11‑22 %, and the closed‑loop workflow (graph compute → database → visualization → business rules → recompute) accelerates investigations from minutes to seconds.

11. Platform Open‑Source and Cloud Services – Angel‑Graph is open‑sourced on GitHub (Angel‑ML) and offered on Tencent Cloud; the graph database product KonisGraph is also available for external users.