Background
When investigating an Entity in the Quantexa Graph, it is common for an investigator to start from the Entity in question, and expand through the Graph along related documents and Entities to find relevant linked information.
This is standard practice in many contexts, for example in various risk applications, through to sanction compliance and identifying marketing opportunities.
In any of these contexts, an investigator may find that they need to make decisions about exactly how to expand through the Graph to make a decision. Expanding just three or four steps outward from an Entity might result in a huge number of Entities and Documents to review (Figure 1). In addition, another Entity or Document’s distance from the Entity in question is often viewed as linked to its relevance.
Because of this, even experienced Quantexa users may be more inclined to limit their expansion of the Graph. But what if this results in important, missed contextual information?
Figure 1. In Graph applications like Quantexa Explorer, investigators often begin from a specific Entity (teal) and expand through the Graph to find important context. However, what if important context is connected, but distant (red)? Each ‘hop’ from the initial Entity introduces more and more information for an investigator to consider, meaning distant, important relationships can be missed.
Using global Graph context
Proximity to a starting Entity isn’t the only way to assess the relevance of nearby Documents and Entities. While manually reviewing hundreds or thousands of linked Entities and Documents may not be feasible, this is an ideal task for Graph Theory and Graph Machine Learning applications.
Quantexa’s Multi-hop Graph Analysis tool is designed to solve this problem by considering a Graph in its entirety, alongside the context of business input. Given a list of ‘seed’ Entities, Multi-hop Graph Analysis can return a configurable, ranked set of Entities for investigation, alongside the Graph context used to find them.
To do this, Multi-hop Graph Analysis takes full advantage of Quantexa’s Knowledge Graph framework, supporting efficient, scalable analytics on Graphs containing many millions of nodes and edges. The Knowledge Graph framework is interfaced with open-source Graph libraries like NetworkX to analyse features related to proximity, paths and many other graph measures. These measures are combined into a custom ‘contextual centrality’, used to highlight Entities of interest related to the seed Entities provided.
Anti-Money Laundering and Multi-hop
Knowledge Graph and Multi-hop Graph Analysis can be used to investigate hidden relationships between those suspected of illicit financial activity, shedding light on undiscovered bad actors (Figure 2).
Figure 2. Identifying potential bad actors with Quantexa and Multi-hop Graph Analysis
To do this, we first take global Dun & Bradstreet corporate registry data, and resolve Entities using Quantexa’s Entity Resolution. The resulting network can be converted into a Knowledge Graph with millions of connected Entities.
Alongside this, we take 13 ‘seed’ businesses suspected of money laundering, identified in the publicly available International Consortium of Investigative Journalists’ (ICIJ) Cyprus Confidential investigations. These investigations explore Cypriot businesses implicated in evading sanctions placed on Russia, following the invasion of Ukraine.
We can run Multi-hop Graph Analysis with these inputs to identify Entities with a sufficiently high ‘contextual centrality’. We find a further 65 newly identified Entities worthy of further investigation, not previously mentioned in ICIJ’s reports. The majority of newly discovered businesses are registered in Cyprus, Ireland and the United Kingdom, revealing the international scope of the suspected money laundering. 19 of these Entities are individuals, around half of which are found on existing watchlists, such as Special Interest or Politically Exposed Persons.
Summary
Quantexa’s Knowledge Graph and Multi-hop Graph Analysis can be used to uncover complex relationships between bad actors. This approach can also be applied to other contexts, wherever there is a need to discover hidden relationships alongside business input.
These results can be combined with Entity and Network Scoring in Quantexa Explorer to aid in complex investigations, allowing investigators to focus in on likely Entities of interest.
To see this example and more in action, view the accompanying recorded webinar.
