Graph Analytics

Graph analytics is a branch of data science that deals with analyzing information networks known as graphs, and extracting information from the data relationships. It ranges from basic measures that characterize graphs, over PageRank, to complex algorithms. Common use cases include fraud detection, recommender systems, and network flow analysis.

ArangoDB offers a feature for running algorithms on your graph data, called Graph Analytics Engines (GAEs). It is included in the Arango Contextual Data Platform.

Key features:

• Separation of storage and compute: GAEs are a solution that lets you run graph analytics independent of your ArangoDB Core, including on dedicated machines optimized for compute tasks. This separation of OLAP and OLTP workloads avoids affecting the performance of the transaction-oriented database systems.

• Fast data loading: You can easily and efficiently import graph data from ArangoDB and export the computed results to a dedicated collection in ArangoDB.

• In-memory processing: All imported data is held and processed in the main memory of the compute machines for very fast execution of graph algorithms such as connected components, label propagation, and PageRank.

Get started

You can interact with Graph Analytics Engines through:

• Web Interface: Control the engines through a graphical user interface. It makes the workflow and management very approachable.

• HTTP API: Programmatically start and stop engines, load data, run algorithms, and store the results. Use it if you want to automate workflows with code.

Note that you cannot interactively explore the graph data that is loaded into an engine, nor inspect the computed values before you store them. Once the results are written to a collection, you can query them with AQL and inspect the result documents.

Available Algorithms

• PageRank: Measures node importance based on incoming connections and their quality.
• Weakly Connected Components (WCC): Find groups of nodes connected by any path (ignoring direction).
• Strongly Connected Components (SCC): Identifies groups where every node can reach every other node via directed paths.
• Betweenness Centrality: Measures how often nodes appear on shortest paths between other nodes.
• LineRank: PageRank applied to edges, measuring the importance of connections.
• Label Propagation: Fast community detection algorithm that propagates labels through the network.
• Attribute Propagation: Propagate and accumulate labels through the graph structure.

See the HTTP API documentation for detailed parameters and usage examples. You can also run algorithms using the web interface.