The app that maps Europe’s online political discourse
The internet’s impact on EU governance and democracy is undisputable. It facilitates the provision of public services, while offering new avenues for political engagement and information. Yet, this evolution presents both challenges and opportunities. “While the internet mobilises supporters, online activities, as seen during recent elections for example, raise concerns,” says Marina Costa Lobo(opens in new window) project coordinator of PolarScopEU(opens in new window). “Research evidences increasing ideological polarisation, echo chambers and misinformation; resulting in greater political segregation.” To help improve the quality of online political communication, the European Research Council(opens in new window) funded PolarScopEU project has developed a ‘breakthrough’ app which maps the political and media environment in each EU country. “We hope to help foster more balanced, transparent, open-minded and well-informed online political discussions, improving the quality of our democracy,” adds Costa Lobo from the Institute of Social Sciences(opens in new window) in Portugal, the project host.
Underpinned by content analysis and automated text analysis
Initially planning to expand the computational framework developed by the MAPLE project, changes to Twitter, coupled with the rise of machine/deep learning, altered PolarScopEU’s approach. Consequently, two work strands(opens in new window) were developed: social media analysis of Bluesky content, producing the Bias Plot and similarity tables, alongside EU politicisation analysis, using MAPLE’s coding frameworks and data. For the first product, posts were collected from specific handles and networks, then fed into automated text analysis tools to identify topics, sentiment and EU mentions. Topic detection relied on a Polexlab(opens in new window) language model, fine-tuned to identify 21 policy topics from the Comparative Agendas Project, while sentiments were detected by another pre-existing cardiffnlp/twitter-xlm-roberta-base-sentiment multilingual model. EU references were identified by MAPLE’s multilingual rule-based detector, based on keyword and abbreviation searches. The aggregated data was then used to assess topic diversity, tone and similarity between accounts, visualised by the Bias Plot. For the second, the EU politicisation product, PolarScopEU created an automated coding framework by combining the aforementioned rule-based EU keyword detection, with fine-tuned classifiers identifying when the EU is central to text, then assigning relevant scores related to issues such as conflict between actors. “Compared to existing studies, we offer a more detailed analysis of EU politicisation, including identifying the EU dimension being discussed – membership, constitutional structure, EU policy etc.” explains project researcher Tiago Casal da Silva. To hone the modelling, topic detection thresholds were manually tested, as short social media posts often lead to conservative predictions, for example frequently underemphasising the presence of policy-related content. Case examples were used to assess category plausibility, especially where posts straddled more than one topic. The EU detection system was tested across several languages, with support implemented for seven. A trigger column was added highlighting which keyword activated the EU flag, which helps identify false positives. “The system works well as a descriptive research tool, although testing highlighted the importance of careful interpretation, along with the need for adequate computing infrastructure,” notes Casal da Silva. Both tools are available within the free fully operational PolarScopEU app(opens in new window).
For anyone interested in EU-related political communication
PolarScopEU has already been adopted for current research, such as analysing Portuguese media articles about the EU(opens in new window), alongside the policy topics and tone. Analytically, the next step is to extend language support, update/refine the current models, and apply the tools to new case studies and comparative settings. Technically, the plan is to improve the handling of long-running analyses – currently constrained by server/hardware limitations. “But we don’t envision a clear end point for this app/project. We are committed to its continuous development,” adds Costa Lobo.
