Networks, narratives, democracy
Computational research on social discourse, polarization, and the health of democratic deliberation.
The challenge
Democratic deliberation requires citizens to encounter perspectives they disagree with, to find points of convergence where they exist, and to distinguish good-faith disagreement from coordinated harassment or manufactured outrage. The recommendation systems, ranking algorithms, and engagement-maximizing feeds of contemporary platforms work against each of those conditions. The result, documented across a now-substantial computational-social-science literature, is a set of observable patterns: rising polarization between user clusters, declining cross-cluster engagement, accelerating diffusion of false or inflammatory content, and the emergence of coordinated networks that exploit those dynamics.
The same digital footprints that platforms use to amplify these dynamics can also be used to measure them — and to design responses. That is what this research line does. Capybara contributes published methods for analysing political convergence on Twitter, network-theoretic characterisations of digital violence, and applications to public-health communication and crisis response. We work with academic collaborators, civil-society organisations, government agencies, and independent media; we publish openly; and we do not work for political campaigns.
The algorithms
Methodological commitments
Computational analysis of social discourse runs into ethical and methodological hazards that don't apply to climate modelling. We commit to the following on every engagement:
| Aggregate-only analysis | All findings reported at cluster, community, or population level. No individual user profiles, no identification of specific accounts (except those that are already public-figure accounts with significant reach). |
| Publicly available data only | No private messages, no purchased platform data, no scraping that violates platform terms. The methods work on public APIs and open archives. |
| Bias declared upfront | Social-media and news data systematically under-represent populations without good internet access, in languages with poor NLP coverage, or in regions with limited press freedom. Every report names which voices the data captures and which it does not. |
| Open methods, peer review | The methods are published in peer-reviewed venues. The code (where the engagement permits) is open. The conclusions can be audited, criticised, and improved. |
Applications
| Crisis-communication design | Real-time monitoring of negative-sentiment spikes, misinformation clusters, and trusted-voice networks during public-health and disaster-response operations. Used in the Bogotá COVID-19 work to guide rapid messaging adjustments. |
| Election-period observatories | The political-convergence framework deployed as an election-period monitoring tool — measuring polarisation, narrative diffusion, coordinated-attack incidents, and the health of cross-cluster engagement across a campaign cycle. Outputs to academic researchers, electoral-observation missions, and civil society. |
| Digital-violence early warning | Anomaly detection on harassment-network growth, applied for partners in government, journalism-protection NGOs, and civil-society digital-rights organisations. Supports rapid response when a coordinated attack on a journalist, activist, or minority community starts to form. |
| Climate-narrative monitoring | Integration with the Capybara climate situation reports: when a wildfire or flood is forecast, the same monitoring stack tracks how the event's narrative is forming, which official sources are being trusted, and where misinformation is gaining traction. |
Who this is for
| Academic researchers | Collaborations on political-network analysis, polarisation measurement, computational sociology, and digital-democracy research. Co-authorships welcome. |
| Civil society & press-freedom organisations | Digital-violence early warning for protection of journalists, activists, indigenous leaders, and minority communities. Misinformation and harassment tracking to support rapid response. |
| Election-observation missions & democracy watchdogs | Election-period monitoring of conversation health, polarisation indices, coordinated-attack incidents, and cross-cluster engagement. Methodologies are open and replicable across electoral contexts. |
| Independent media & investigative journalism | Data partnerships for investigative reporting on coordinated disinformation, electoral interference, harassment networks, and digital-rights violations. Findings published with full methodology. |
What we don't do
| Work for political campaigns | We do not provide audience targeting, message amplification, or sentiment-optimisation services for candidates, parties, or campaign committees. The methods could be misused this way; we don't allow it. |
| Identify or profile individual users | Even when accounts behave clearly as coordinated bots, our reporting describes the cluster, not the account. We do not produce target lists of any kind. |
| Surveillance work | We decline engagements from governments or agencies whose primary use case is monitoring activists, journalists, dissidents, or minority communities — regardless of the framing. |
| Publish findings that could enable harassment | We do not publish analyses identifying victim communities, individual harassment targets, or vulnerable populations in ways that could amplify attack vectors against them. |
| Use scraped private data | Public APIs and open archives only. No purchased datasets of dubious provenance, no scraping that violates platform terms, no private messages. |
Partner with us
We work with academic collaborators, civil-society organisations, electoral-observation missions, independent media, and government agencies (within the bright lines above). Typical engagement scopes:
- An election-period monitoring observatory with weekly briefings.
- A digital-violence early-warning system for a press-freedom or human-rights organisation.
- A peer-reviewed research collaboration applying the political-convergence framework to a new electoral context.
- A crisis-communication monitoring deployment paired with a Capybara climate situation report.
Capybara Center · Networks, narratives, democracy · Aggregate-only · Public data · Open methods
Note: Only a member of this blog may post a comment.