Quantifying Public Political Discourse

AgoraLens applies structured language analysis to transform unstructured public communication into consistent, comparable analytical indicators.

The platform is designed to help analysts observe how public political discourse evolves over time — not to judge intent, beliefs, legality, or personal character.

1. Data ingestion

AgoraLens continuously monitors public, on-record accounts of elected officials and key political figures using supported platform APIs.

Scope

Public accounts of Finnish Members of Parliament and major party leadership.

Update cadence

New public posts are detected and processed continuously, with typical update intervals of approximately 15 minutes.

Privacy boundaries

  • Only publicly available content is analyzed
  • No private messages, protected accounts, or non-public data are accessed

2. Linguistic analysis

Each new public post is processed using structured language analysis models designed to identify observable rhetorical features.

These models evaluate characteristics of language such as:

  • framing intensity,
  • use of generalized or abstract group references,
  • emotionally charged or escalating phrasing,
  • presence of conciliatory or constructive language.

Rhetorical Intensity Index (0–100)

The Rhetorical Intensity Index is a composite indicator summarizing the overall rhetorical force of a public statement based on weighted linguistic features.

Example feature categories

  • Abstract or dehumanizing group references (higher weight)
  • Confrontational or violent imagery (higher weight)
  • Broad group generalizations (medium weight)
  • Explicitly constructive or conciliatory language (reduces score)

All scores are contextual indicators, not factual claims or judgments.

3. Audience reaction and response dynamics

AgoraLens also analyzes how audiences respond to public statements over time.

Audience reaction classification

In addition to response volume and timing patterns, AgoraLens analyzes the content of public replies to identify how audiences are reacting to specific statements.

Reaction analysis is based on a representative sample of public replies and focuses on observable patterns, including:

  • supportive responses,
  • oppositional responses,
  • mixed or unclear reactions,
  • and instances of audience backlash.

Backlash is defined as a statistically notable concentration of negative or oppositional reactions relative to the baseline response pattern for similar posts.

Reaction labels describe observable reply patterns only. They do not assert intent, coordination, motivation, or legitimacy of the reactions.

Deterministic classification logic

Reaction classification combines structured language interpretation with deterministic thresholding to ensure consistency across time and actors. While individual linguistic features are probabilistic, the assignment of reaction categories (e.g. backlash detected vs not detected) follows explicit, versioned rules.

Response volatility

Rather than focusing on absolute engagement volume alone, we examine both audience reaction composition and response volatility — changes in how and when reactions occur following specific posts.

Response volatility indicators may reflect:

  • polarization of replies,
  • sudden amplification or drop-off in engagement,
  • clustering of reactions within short time windows.

These patterns describe audience behavior, not endorsement or opposition by AgoraLens.

In low-volume contexts, reaction analysis may be marked as preliminary or unavailable when insufficient reply data is present. These cases are explicitly labeled to prevent over-interpretation.

4. Ethics, limitations, and oversight

Automated analysis of language is inherently imperfect.

AgoraLens is designed as a decision-support tool, not an automated judge.

Safeguards include:

  • clear separation between indicators and interpretation,
  • emphasis on trends rather than individual posts,
  • human review of aggregate findings used in analytical briefings or reports,
  • continuous refinement of definitions based on academic and expert feedback,
  • no claims about coordination, manipulation, or intent behind audience reactions.

All metrics are probabilistic and should be interpreted alongside original public content and contextual knowledge.

5. Transparency commitment

To support transparency and responsible use, we publish:

  • metric definitions and category boundaries,
  • high-level weighting and thresholding principles,
  • update notes when analytical models, thresholds, or definitions are adjusted.

This transparency allows users to understand how results are produced and where their limitations lie.

6. Feature functionality explained

This section details the specific logic and criteria behind key platform features.

High Risk Alerts

The High Risk Feed automatically surfaces posts that exhibit elevated rhetorical intensity combined with specific risk markers. It is designed to help analysts quickly identify statements that may signal escalation.

Trigger Criteria

A post appears in the High Risk Feed if it meets one or more of the following conditions:

  • High Intensity: Rhetorical Intensity Index score ≥ 70/100.
  • Significant Volatility: Audience Response Volatility score ≥ 70/100.
  • Escalation Markers: Presence of specific linguistic markers such as explicit confrontations, dehumanizing language, or violent imagery (regardless of total score).

Note: Not all high-risk alerts represent harmful or policy-violating content. A "High Risk" label indicates structural linguistic escalation that warrants analyst attention, not necessarily a judgment of the content's validity.

Analysis Snapshot

The Analysis Snapshot on individual post pages provides a frozen view of analytical metrics at a specific moment in time. Because social media content is dynamic, scores are anchored to a "Snapshot Window" to ensure consistency in reporting.

Key Metrics Explained

Engagement Velocity

This histogram visualizes the rate of audience interactions over time. Peaks indicate moments of viral acceleration or coordinated response bursts.

Backlash Labeling

If the pattern of replies meets the deterministic criteria for Audience Backlash (statistically high concentration of negative/oppositional language), a warning label appears in the snapshot. This serves as a structural indicator of audience reception, distinct from the post's content.

Note: The snapshot captures the state of the post at the time of analysis. If a post is deleted or significantly altered later, the snapshot serves as a historical record of its state during the active monitoring window.

Metric Deep Dive

The Metric Deep Dive widget breaks down the high-level Rhetorical Intensity score into its constituent linguistic components. These sub-scores help verify why a particular intensity level was detected.

Sub-Score Categories

Confrontational Imagery

Detects language that evokes conflict, physical struggle, or militaristic framing. High scores correlate with rhetoric that frames politics as a battle or existential fight.

Abstract Group Reference

Identifies references to "Them," "The Elite," "Globalists," or other generalized out-groups. This is a key marker of populist or polarizing rhetoric, distinct from specific criticism of named individuals.

Broad Attribution

Measures the extent to which negative traits or actions are ascribed to entire groups rather than individuals. High scores often signal stereotyping or collective blame.

Conciliatory Language (Mitigating Factor)

Detects attempts to bridge divides, acknowledge nuance, or de-escalate. Unlike the other metrics, a high score here reduces the overall Rhetorical Intensity.