Problem Addressed
Local Facebook groups are often a rich source of information for campaigners, providing insight into the issues, concerns and conversations shaping communities. While campaign teams frequently monitor individual posts and comments, doing so at scale is difficult. We wanted to explore whether AI-assisted social listening could help identify the issues people are discussing across different parts of the country and provide campaigners with a more systematic understanding of local political conversations.
Approach & Implementation
We collected publicly available discussions from over 1,000 local Facebook groups across the UK and processed the data using Categorum, our AI-powered classification system.
The first stage of analysis filtered out non-political content, allowing us to focus on discussions relevant to public policy, local services, elections and political issues. We then used a identified recurring themes and classified conversations into issue categories and subcategories. Rather than relying on a fixed coding framework, the model generated categories from the data itself, helping surface emerging issues and local concerns.
Once classified, discussions were mapped geographically, allowing us to analyse which issues were most prominent in different areas and compare patterns of concern across constituencies and regions.
Because political discussions are often complex and overlap multiple topics, categories should be understood as indicators of dominant themes rather than rigid labels. AI classifications were tested against samples reviewed by human volunteers to ensure that the outputs were broadly accurate and useful for campaign analysis.
As is the case for all of our tools, iterative improvement is only made possible by testing and user feedback. If you have any feedback for this tool, please feel free to get in touch with [email protected]