What is graph traversal?
Graph traversal is the process of exploring connected data by following the relationships between entities to uncover patterns, detect cycles, find shortest paths, and more.
In an intelligence investigation, those entities might include people, organisations, vehicles, locations, devices, financial accounts or events. Rather than viewing each record in isolation, graph traversal allows analysts to move naturally from one entity to the next, uncovering connections that help explain the wider picture.
Every time an analyst expands a person’s associates, follows a chain of financial transactions or traces ownership through multiple companies, they are performing a graph traversal.
Although graph traversal is a fundamental capability of graph databases, analysts rarely think about the technology behind it. Instead, it becomes part of the investigative workflow, allowing them to ask new questions, explore emerging hypotheses and follow evidence wherever it leads.
What is a graph?
A graph is a way of representing connected data. It consists of nodes, which represent entities such as people, organisations, vehicles, locations or financial accounts, and relationships, which describe how those entities are connected.
Unlike traditional tables, graphs are designed to model real-world relationships directly. This makes them particularly well suited to intelligence analysis, where understanding how entities connect is often just as important as understanding the entities themselves.
Graph traversal is the process of navigating these relationships, allowing analysts to move naturally from one entity to the next as an investigation develops.
Why graph traversal matters
Most investigations begin with only a small amount of information.
An analyst might start with a telephone number recovered from a mobile device, the registration number of a suspicious vehicle, a company director identified during a financial investigation or a single suspicious transaction. The challenge is understanding what else is connected.
Graph traversal allows analysts to move through the surrounding network one relationship at a time.
For example, they might discover:
- who owns the telephone number
- other devices linked to the same individual
- companies where that individual is a director
- vehicles registered to those companies
- properties associated with those vehicles
- financial accounts linked to those properties
- other investigations involving the same entities
Each step provides additional context, helping analysts build a richer understanding of the investigation.
Graph traversal and intelligence analysis
Graph traversal is not an analytical technique in its own right.
Instead, it provides the mechanism that supports many of the activities intelligence analysts perform every day.
For example:
- expanding networks during link analysis
- tracing beneficial ownership
- following communication chains
- identifying indirect relationships
- exploring organised crime groups
- understanding financial flows
- investigating hostile networks
Graph traversal makes these investigations interactive, allowing analysts to explore connected intelligence as new information becomes available.
Multi-hop graph exploration
Many of the most valuable investigative insights are not found in direct relationships. Instead, they emerge several steps away from the original investigative lead.
For example, a suspect may not own a company directly, but may be connected through a complex relationship:
Suspect → Associate → Director → Company → Property
Similarly, two individuals who appear unrelated may be linked through shared addresses, vehicles, financial transactions or previous investigations. Graph traversal allows analysts to follow these indirect, or multi-hop, relationships without needing to define every possible path in advance.
This flexibility makes graph technology particularly valuable when investigating organised crime, national security threats and complex financial crime.
Graph traversal vs traditional database queries
Traditional databases excel at retrieving known information.
For example:
- Find this customer.
- Retrieve this transaction.
- List these case records.
Investigations are rarely so straightforward.
Analysts often begin with one question, only to discover several new ones as they explore the available information. Graph traversal supports this style of investigation by allowing analysts to follow relationships dynamically rather than relying on predefined joins or fixed reporting structures.
Instead of asking a single question, analysts can continuously refine their investigation as new connections emerge.
Graph traversal and graph databases
Graph traversal is possible using many different technologies, but graph databases are specifically designed to make traversing connected data fast and efficient.
Rather than repeatedly joining tables to reconstruct relationships, graph databases store those relationships directly, allowing analysts to explore complex networks interactively.
As investigations grow in complexity, this becomes increasingly important.
Analysts can continue exploring large, highly connected datasets while maintaining the responsive, interactive experience needed for investigative work.
Graph traversal and GraphRAG
Modern AI systems increasingly combine graph traversal with large language models. Within a GraphRAG architecture, graph traversal helps identify the entities, documents and relationships that provide relevant context before information is passed to the language model.
For example, an analyst might ask “show me every company indirectly owned by this individual.”
Behind the scenes, the platform performs multiple graph traversals to identify connected entities and retrieve supporting evidence before generating a response.
This enables more accurate, explainable and context-aware answers than searching disconnected documents alone.
How GraphAware Hume uses graph traversal
Graph traversal underpins almost every investigative workflow within GraphAware Hume.
When analysts expand a network, search for connected entities, switch between graph, map and timeline views, or investigate indirect relationships, Hume performs graph traversals behind the scenes.
The analyst does not need to understand graph algorithms or write complex queries. Instead, they can focus on the investigation itself, exploring connected intelligence naturally while Hume retrieves the relevant relationships in real time.
Combined with entity resolution, graph visualisation, and AI-assisted investigation, graph traversal helps analysts move from isolated records to a connected intelligence picture more quickly and with greater confidence.
Frequently asked questions
What is graph traversal?
Graph traversal is the process of exploring connected data by following relationships between entities such as people, organisations, locations, assets and events.
Is graph traversal the same as link analysis?
No. Link analysis is an investigative technique used to understand relationships within connected data. Graph traversal is the mechanism that allows analysts to explore those relationships interactively.
Does graph traversal require a graph database?
Not necessarily. However, graph databases are specifically designed to traverse connected data efficiently, making them particularly well suited to investigations involving complex, multi-hop relationships.
What is a multi-hop traversal?
A multi-hop traversal follows relationships across several connected entities rather than stopping at direct neighbours. Many important investigative insights emerge through these indirect connections.
How does graph traversal support AI?
Modern GraphRAG and agentic AI systems use graph traversal to retrieve relevant entities, relationships and supporting evidence before generating responses. This helps produce more accurate and explainable outputs grounded in connected data.
Do analysts need to understand graph algorithms?
No. Modern intelligence platforms perform graph traversals automatically, allowing analysts to explore connected data through intuitive visual interfaces and natural language queries rather than needing to understand the underlying algorithms.