Seeing Migration Narratives: Enhancing Decision Making Through Visualisation of Policy Debates for Civil Society and the Media
Lead Research Organisation:
University of Oxford
Department Name: Oxford Internet Institute
Abstract
Migration and refugee issues are among the most complex and polarising policy challenges in the UK and globally. Media narratives on migration are perceived to have a significant impact on policy decisions and public attitudes toward migration. However the enormous number of media stories about migration, and the immense complexity of their diffusion around digital platforms make it impossible, using existing tools, to track and understand the relationship between different narratives, actors and sources in the debate. This creates space for the emergence of one-sided and misleading public debates, which can be exploited to generate discord and fear and lead to policy making that serves neither the national interest nor the interests of migrant and refugee communities.
This project will develop and apply novel techniques from natural language processing, machine learning, statistics, and data visualisation to: reveal the content and nature of digital discourse on migration; track the diffusion of different stories and concepts; demonstrate the reach of different voices in the debate. This will provide clear evidence of the sources of power and influence within the debate, and allow civil society actors, policy makers and journalists dealing with the issue to make informed decisions on what action may be taken to hold this power to account.
This project will develop and apply novel techniques from natural language processing, machine learning, statistics, and data visualisation to: reveal the content and nature of digital discourse on migration; track the diffusion of different stories and concepts; demonstrate the reach of different voices in the debate. This will provide clear evidence of the sources of power and influence within the debate, and allow civil society actors, policy makers and journalists dealing with the issue to make informed decisions on what action may be taken to hold this power to account.
Publications
Hale S
(2024)
Analyzing Misinformation Claims During the 2022 Brazilian General Election on WhatsApp, Twitter, and Kwai
in International Journal of Public Opinion Research
Shliselberg M
(2024)
SynDy: Syn thetic Dy namic Dataset Generation Framework for Misinformation Tasks
| Description | Submitted Evidence to Parliament: Science, Innovation, and Technology Committee |
| Geographic Reach | National |
| Policy Influence Type | Contribution to a national consultation/review |
| Description | Meedan Timpani and Presto |
| Organisation | Meedan |
| Country | United States |
| Sector | Private |
| PI Contribution | Meedan, a global non-profit, has spent over $100k USD building open-source software for collecting and anlaysing social media data. Meedan would also like to identify meso- and macro-level narratives in their data similar to our project (although in foci other than migration). We are examining the infrastructure they have built and understanding if it might be a suitable foundation upon which to build the service we ultimately want to make available to NGO's in the UK. |
| Collaborator Contribution | Software to host machine learning models in a scalable manner, consume social media feeds, cluster social media content using semantic similarity models, and visualize the clusters in an interactive way. |
| Impact | TBD |
| Start Year | 2023 |
| Title | Presto |
| Description | Presto is a Python service that aims to perform, most generally, on-demand media fingerprinting and other machine learning tasks at scale. In the context of text, fingerprints are transformer vectors - video is done by TMK, images by PDQ, and audio by chromaprint. The service also hosts keyword extraction and categorization models. |
| Type Of Technology | Webtool/Application |
| Year Produced | 2024 |
| Open Source License? | Yes |
| Impact | Presto provides a scalable infrastructure to easily host machine learning models so that they can be used by other services such as Timpani (reported separately) |
| URL | https://github.com/meedan/presto |
| Title | Timpani Data Analysis |
| Description | Timpani is a flexible machine learning infrastructure for managing data collection, transformation, modeling and inference. Timpani has tools for downloading content items from API or other datasource, processing through various models (clustering, keyword extraction, etc), and a viewer for exploring trends in content. |
| Type Of Technology | Webtool/Application |
| Year Produced | 2025 |
| Open Source License? | Yes |
| Impact | The development of Timpani in cooperation with Meedan, a charity working with fact-checkers and journalists, has allowed us to prototype several elements of narrative extraction. These include building semantic clusters from large datasets, extracting keywords to discover new trends, and using generative AI to categorize social media posts into a user-supplied taxonomy of narratives. |
| URL | https://github.com/meedan/timpani-public |
| Description | London migration co-creation workshop |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Third sector organisations |
| Results and Impact | This workshop with civil society organizations working in roles related to migration focused on gathering an understanding of (1) the needs the organizations had, (2) their understandings of what a 'narrative' is and what narratives about migration are most prevalent in the UK, and (3) how technology could help meet the unmet needs of these organizations. |
| Year(s) Of Engagement Activity | 2025 |
