Using Large Language Models to support Government Policy & Programme Delivery

Background

In May 2021, the Department for the Economy published its 10X Economy – an Economic Vision for a Decade of Innovation Report. This 10X Economic Vision sets out an ambitious strategic vision for innovation led, inclusive and sustainable economic growth in Northern Ireland.

To help support the delivery of this vision, the Department for the Economy created an SBRIproject designed to identify those programmes and policies that are relevant to 10X on an ongoing and real time basis.

As part of the project, the Department for the Economy considered it vital that relevant performance indicators associated with those policies and programmes are captured in a comprehensive and coherent manner. The project also sought identify where weaknesses existed and what gaps needed to be filled, accounting for supporting strategies and linked policies that continue to evolve over time.

The solution

In response, Analytics Engines created a solution that quickly and easily provides rich and meaningful insights into a text base. Utilising emerging Large Language Models technologies, the solution is comprised of four main components, each addressing a specific requirement of the Department for the Economy.

• Advanced Search
  Utilising Large Language Models, the solution has a computational understanding of the topics being discussed; enabling users to search the text base for contextually relevant, semantic insights. In addition to returning source text, the advanced search feature presents users with a Large Language Model generated response that details how their input query is being discussed across all documents.
• Topic Extraction
  The solution analyses the entire text base, providing users with insights into commonly discussed topics and concepts. Users are presented with a Large Language Model generated summary about a given subject. In addition, sentiment analysis provides users within insights into how extracted topics are being discussed, providing positive, neutral, and negative ratings for each.
• Structured Data Extraction
  The solution is able to identify and present structured data such as names, dates, events, and organisations from raw text and documents. These entities can be defined and customised by the user, enabling the extraction of critical information and insights.
• Network Graph
  Utilising Knowledge Graph technologies, the solution presents users with a network graph that shows the semantic relationship that exist between individual documents in the text base, enabling users to identify and explore relevant documentation more effectively.

In addition, and to support the effective monitoring of relevant key performance indicators, the solution features a comprehensive data visualisation dashboard, which pulls together publicly reported data on key metrics of interest, demonstrating the effectiveness and impact of individual policies.

The impact

The solution supports the automated discovery and extraction of relevant insights and information in support of the Department for the Economy’s ambitious innovation roadmap. The solution enables the organisation to analyse documentation in a way that would not have been previously possible, significantly reducing time to insight, and driving greater operational efficiency.