The challenge
GraphAware’s client is a government agency dedicated to protecting the country by combating serious and organised crime. As a statutory authority, the law enforcement agency governs itself, using a complex framework to dictate how the Agency should operate and conduct its business.
The governance framework consists of different types of governance instruments, ranging from publicly available legislation, commissioner’s orders, and case law to internal directives, delegations, enterprise and employment agreements.
With nearly a thousand different instruments totalling tens of thousands of pages, any organisational or legislative change creates a significant challenge. It takes a huge human effort to ensure that the legal obligations placed on the organisation are followed correctly and consistently, and that it is always clear who is accountable and responsible for these obligations.
The Agency’s commitment to continuous improvement only amplifies the need to keep the documentation up to date.
The engagement with GraphAware has been first class. There’s been excellent access to experts and expertise, incredible support and response times, and a high quality product that’s on the mark.— Lacey Barnes, Agency Architect
The solution
Lacey Barnes is an Architect at the Agency. Having previously implemented a graph-powered approach for operational use cases, Lacey had already experienced the benefits of making hidden connections explicit. By creating a single connected view of truth from siloed, disconnected, and heterogeneous data, she could empower teams of analysts and investigators to work more efficiently and effectively.
Most of the documentation is in the public domain, using hyperlinks to refer to pieces of legislation and other instruments. Sketching out the governance problem domain, Lacey quickly realised that the entities and relationships amongst them naturally form a graph.
Other kinds of structured data were also available to enrich the network of interlinked documents. Organisational units, position hierarchies, and the names and roles of current and historic agency staff could be harvested from the Agency’s enterprise resource planning (ERP) system.
Lacey’s team began building a governance knowledge graph from the structured data at hand, with the intention of exposing it to end users through GraphAware Hume’s powerful graph-powered search and exploration interface.
“If we can help the business easily explore connections between instruments, identify documents with a high degree of centrality or clusters of similar documents, it will already be a huge win” said Lacey.
She hoped to alleviate parts of the manual process by automating change impact analysis, prioritising the review of densely connected documents, and detecting potential duplicates.
Lacey asked the GraphAware team to help leverage Hume’s natural language process (NLP) capabilities to automatically detect mentions of other governance instruments, positions, keywords, obligations, as well as other entities and semantic relationships among them.
After two months of close collaboration between GraphAware and the Agency’s data science teams and domain experts, the machine learning models were trained and finetuned, ready to be plugged into Orchestra, Hume’s data ingestion and enrichment workflow engine.
Here, Orchestra was used to:
- Ingest documents from various locations
- Perform optical character recognition (OCR) if needed
- Enrich the data using machine learning algorithms
- Perform necessary postprocessing and deduplication
- Convert the data into the right knowledge graph structure
- And finally store it in Neo4j, the underlying graph database.
With a full governance knowledge graph now in place, the Agency teams created a set of preconfigured traversals through the network – called Hume Actions – which allow end users to visually interact with the graph and ask insightful questions.
For example, they can search for a piece of legislation that has recently changed and learn, with a single click, which governance instruments may be impacted by this change and why. Similarly, end users can find out which governance instruments refer to positions that no longer exist in a matter of seconds.
The results
By implementing a governance knowledge graph using GraphAware Hume and exposing it to non-technical end users and domain experts, the Agency reduced the time required to analyse the impact of a legislation change from days to minutes.
GraphAware Hume automatically identifies gaps, duplicated documentation, missing obligation implementations, and even recommends the best way to resolve outdated references to the organisational structure.
It has significantly reduced the risk of the agency falling out of compliance, and freed-up valuable time previously spent manually sifting through thousands of pages of documentation. The Agency’s staff is better empowered to proactively execute on the Agency’s commitment to continuous improvement and better self-governance.
“The engagement with GraphAware has been first class. There’s been excellent access to experts and expertise, incredible support and response times, and a high-quality product that’s on the mark.” said Lacey Barnes, Architect at the agency. “Issues and challenges have been resolved in 24 hours consistently”, she added.
As a result of the governance knowledge graph project going live, other teams within the agency have realised the potential of GraphAware Hume and Neo4j for additional use cases and, ultimately, for the organisation’s mission to protect its country’s people interests.