To get the most insight out of your data, you need to add a little math. Math in the form of graphing technology, that is. Graphing not in the sense of visualizations – although that’s part of it – but graphing in the sense of discovering the relationships among data elements such that you get insights otherwise hidden.

John Bender, the regional vice president for U.S. federal at Neo4j, says data users must understand the difference between relational and graphing database technology.

“A relational database uses tables and then it uses joins to bring the data together,” Bender said. “Whereas a graph looks at the data and the relationships as equal. So we can query off anything, any relationship or any node in the graph.”

For how this benefits users, Bender used the example of fraud detection, something at the top of mind for many federal IT practitioners. By better understanding the network of entities revealed by a graphing database, evidence of questionable activity will be more visible.

“If someone’s doing money laundering,” Bender said, “you’re trying to understand the network [and] what kind of activities they are doing for the evasion.” Known commonalities of various types of fraud form clusters under the right algorithmic queries, he said.

Bender said that the more sources of data brought into the graph database, the more, and more clearly seen, commonalities emerge.

“You’re going to find as you start bringing in other sources, they have commonality in the ‘nouns’ or in the nodes, so new relationships start showing up,” Bender said.

Technically, the agency would retain the databases tied to specific applications, even while contributing them to the graph database.

“We’re not replacing the silos,” Bender said. “They’re purpose built for those applications.” Instead, the silos of data are combined in a data lake, “and that’s where the graph goes.”

An army of data points

In one instance, the Army is using Neo4j to better understand and manage its supply chains. In moving around platforms made of hundreds of thousands of parts, “they had a hard time making sure that had the right amounts of parts,” Bender said, “because they didn’t have deep understanding of that their need was.” Logisticians would order months in advance and stockpile, just in case.

Now, Bender said, the Army has a graph database with 8 million nodes and 21 million relationships to help supply officers model conditions, run what-if scenarios and generally improve readiness.

Law enforcement and health care benefits management also make good use cases for graph technology, Bender said.

One federal agency is using it to improve the claims process in a couple of ways.

First, “they’re using the graph to find a way to make it better for the citizens, using graph to understand their journey so the agency can understand ways to improve it,” he said. “They’re also analyzing claims and looking for anomalies and how people are submitting the claims and looking for the patterns of fraud.”

In another use case, a contractor built a digital twin of an agency network. By graphing the relationships among network nodes and simulating intrusions, the client can improve cybersecurity.

You can “start to look for the combination of events and relationships and similarities to other events,” Bender said. “If this event happened, I can look at the other events that had a certain settings. I can take that and use it as a search parameter to other events, looking for other things that match.”

In another context, Bender said, NASA has built a lessons-learned graph database, ingesting decades of documentation of projects and experiments. Bender said it will help the agency avoid duplication of prior efforts while helping enrich current and future ones.

And because graph database technology scales so easily, Bender said, “you can continually add more nodes as you go. You don’t have to stop and rebuild the whole graph. So as agencies start to learn more, as they become more accomplished with their graphs, they start asking more and more connected questions.”

Bender said Neo4j is architected for speed, with the graph intelligence able to run from RAM (random-access memory). The product can run on agency premises or in the cloud, or in a combination of both.

Wherever it runs, Bender noted, graph technology, coupled with retrieval augmented generation, can greatly refine results when applied to publicly-trained large language models. It helps the user obtain an output using specific data, rather than everything that might have trained the LLM.

“We can actually take a look at a generative AI app before the actual query or the question,” Bender said. The resulting answer “will come through the [agency’s own] graph using the inside data.” He added, “The whole goal now is to make information more accurate.”

Listen to the full show:

Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.