Social analytics for information integrity review

Vectocrat helps analysts inspect suspicious short-video account behaviour with explainable evidence.

The product combines multimodal content enrichment, behavioural baselines, compound anomaly signals, and account-level explainability so investigators can move from broad monitoring to specific cases.

Open demo See methods

Method stack

The platform is built around four method layers.

Vectocrat OÜ does not rely on a single score. It joins content understanding, time-series baselines, signal combination, and analyst-facing decomposition so unusual behaviour can be reviewed in context.

Content enrichment

Post metadata is combined with inferred topic, sentiment, stance, disclosure, and campaign-related variables so the system can reason about meaning as well as raw volume.

Behavioural baselines

Weekly account metrics are compared against recent account history and cohort behaviour. The dashboard then shows observed values, recent baselines, and anomaly markers side by side.

Compound anomaly signals

Follower spikes, disclosure drops, engagement distortions, and stance shifts stay weak on their own. They become more meaningful when multiple signals align in the same period or sequence.

Explainable review

Each flagged account can be opened into component-level evidence with baseline, observed value, affected period, and narrative explanation instead of a black-box label.

Analyst workflow

From monitoring to case review.

Filter accounts by topic, sentiment, platform, segment, and anomaly signal.
Check observed-vs-baseline trend movement for the active slice of accounts.
Rank accounts by anomaly score and inspect their strongest contributing signals.
Open a case to review topic breakdowns, time-series detail, and score-component evidence.

Weekly time-series chart used in the dashboard — Observed metrics, recent baselines, and anomaly markers are presented together for human review.

Feature correlation heatmap — Feature relationships help validate that the score is grounded in behavioural structure rather than isolated spikes.

Anomaly detection metrics chart — Validation views keep the demo tied to measurable detection performance and feature priority checks.

Team

A research and build team spanning computational social science, engineering, quantitative modelling, and media design.

The notes below are based on public institutional and project pages. Where those pages are brief, the role descriptions stay conservative and close to the published source material.

Andres Karjus

Computational linguist and senior research fellow at Tallinn University's School of Humanities. Public CUDAN material describes his work as combining machine learning on large text corpora, computational simulations, and human communication experiments to study language and culture change.

Martin Karjus

ICT specialist at Tallinn University's School of Humanities. Tallinn University pages and EXU project notes place him in technical implementation and research-support work, including contributions to funded AI-based media-monitoring prototype efforts and knowledge-transfer projects.

Sander Paekivi

Physicist and researcher-developer at Tallinn University, with prior postdoctoral work at the Max Planck Institute. His academic background includes modelling complex systems, including diffusion and neural networks, with applications in language group competition and media analytics. That profile fits the rigorous mathematical foundation behind the project's AI systems and large-scale signal analysis.

Ellu-Marie Meos

Media integration and design specialist at Tallinn University's School of Humanities. Public university and EXU pages associate her with media integration, design, and knowledge-transfer initiatives, which fits the product-facing communication and presentation layer of the project.

Get In Contact With Us

Talk to us about solutions, live demos, and media analytics needs.

If you need a tailored monitoring workflow, a live walkthrough of the platform, or support on media analytics and research applications, reach out directly. We can continue from a short intro note and move to a working demo.

Solutions and research

Discuss platform use cases, computational social science methods, and research-oriented analytics work.

Email Andres Karjus

Live demo and technical setup

Request a live product walkthrough or talk through implementation details, prototypes, and technical integration.

Email Martin Karjus

Media analytics collaboration

Get in touch for presentation, communication, and applied media analytics collaboration needs.

Email Ellu-Marie Meos

Use the dashboard demo to inspect the method in action.

The demo keeps the explainability surfaces intact: filters, time-series context, anomaly composition, account detail, and methodology notes.

Launch demo