Crisis Detection & Alerts

Overview

SENTi-radar’s crisis detection system continuously monitors topics for negative sentiment spikes, emotion anomalies, and volatility surges. When a crisis is detected, the platform triggers real-time alerts and elevates the topic to your Crisis & Alerts Log.

Crisis Levels

Every topic is assigned one of four crisis levels:

None

0-1 negative signalsSentiment is stable or positive. Normal monitoring recommended.

Low

2-3 negative signalsMinor concerns detected. Watch closely but no immediate action required.

Medium

4-5 negative signalsElevated concern. Negative emotions are rising. Proactive engagement advised.

High

6+ negative signalsCrisis risk. Immediate attention required. Public anger/fear dominating.

Detection Algorithm

The crisis level is calculated based on negative keyword density:

// Source: TopicDetail.tsx (lines 262-268)
const negKw = [
  'war', 'attack', 'crisis', 'shortage', 'tension', 'conflict',
  'scandal', 'ban', 'protest', 'threat', 'crash', 'decline',
  'fail', 'corrupt', 'dangerous'
];

const negCount = negKw.filter(w => text.includes(w)).length;

const crisisLevel: 'none' | 'medium' | 'high' = 
  negCount >= 4 ? 'high' 
  : negCount >= 2 ? 'medium' 
  : 'none';

The algorithm combines keyword matching with emotion analysis. A topic with 48% anger + 22% fear will typically trigger a “high” crisis level.

Real-World Crisis Examples

Example 1: Climate Summit (High Crisis)

// Source: mockData.ts
{
  title: 'Climate Summit 2026',
  sentiment: 'negative',
  crisisLevel: 'high',
  volatility: 85,
  emotions: [
    { emotion: 'anger', percentage: 42 },
    { emotion: 'sadness', percentage: 25 },
    { emotion: 'fear', percentage: 18 }
  ],
  summary: '42% anger, 25% sadness → Key frustrations: broken promises (55%), insufficient targets (30%), greenwashing (12%)'
}

Why it’s a crisis:

67% negative emotions (anger + sadness + fear)
Keywords: “broken promises”, “greenwashing”, “too little too late”
Volatility: 85/100
Volume spiked 60% after leaked draft targets

Example 2: Food Prices (High Crisis)

{
  title: 'Global Food Prices',
  hashtag: '#FoodCrisis',
  sentiment: 'negative',
  crisisLevel: 'high',
  volatility: 91,
  emotions: [
    { emotion: 'anger', percentage: 48 },
    { emotion: 'fear', percentage: 22 }
  ],
  summary: 'Outrage at: corporate profiteering (52%), government inaction (33%), supply chain failures (12%)'
}

Why it’s a crisis:

70% negative emotions (anger + fear + sadness)
Personal impact: “Can’t feed my family” posts up 200%
Viral evidence: Grocery receipt photos spreading
Keywords: “corporate greed”, “can’t afford”, “price controls”

Example 3: AI Regulation (Medium Crisis)

{
  title: 'AI Regulation Debate',
  sentiment: 'mixed',
  crisisLevel: 'medium',
  volatility: 72,
  emotions: [
    { emotion: 'fear', percentage: 35 },
    { emotion: 'anger', percentage: 28 },
    { emotion: 'surprise', percentage: 18 }
  ]
}

Why it’s medium:

63% negative/anxious emotions (fear + anger)
Polarized debate (not universal outrage)
Main concerns: job displacement (48%), lack of oversight (31%)
Moderate volatility indicates ongoing debate, not crisis

Crisis & Alerts Log

The Crisis & Alerts Log displays real-time notifications for active crises:

Alert Types
UI Display

Negative Spike Detected (🔴 Red)

Triggered when:

Sentiment drops below -20
Anger or fear exceeds 40%
Volume spikes 50%+ with negative tone

Example:

🔴 Negative spike detected
Anger dominant in public discourse
Just now

Sentiment Shift (🟡 Yellow)

Triggered when:

Sentiment changes 30+ points within 6 hours
Volume increases 100%+ (viral momentum)

Example:

🟡 Sentiment Shift
Sudden increase in conversation volume
30 min ago

Stable / Positive (🔵 Blue)

Default state when no crisis detected:

🔵 No critical alerts
Public conversation is moderate
2 hours ago

Each alert includes:

Status dot: Animated pulse for active alerts
Alert title: Clear, actionable label
Description: Specific trigger (emotion name, percentage)
Timestamp: Relative time (“Just now”, “30 min ago”)
Border color: Matches severity (red/yellow/blue)

// Source: TopicDetail.tsx (lines 906-913)
<div className="flex items-start gap-3 p-3 bg-card 
                border-l-2 border-destructive rounded-r shadow-sm">
  <span className="h-2 w-2 mt-1.5 rounded-full 
                   bg-destructive animate-pulse" />
  <div className="flex-1">
    <p className="text-xs font-semibold">Negative spike detected</p>
    <p className="text-[11px] text-muted-foreground">
      <strong className="text-destructive">Anger</strong> dominant
    </p>
  </div>
</div>

Scheduled Monitoring & Alerts

Set up automated crisis monitoring with the Schedule Analysis feature:

Open Schedule Modal

Click the “Schedule Analysis” button in the dashboard header or topic detail panel.

Configure Topic & Frequency

Select the topic to monitor and choose a frequency:

Every hour
Every 6 hours
Daily
Weekly
Custom (cron expression)

Set Alert Thresholds

Define crisis triggers using interactive sliders:

// Source: ScheduleAnalysisModal.tsx
{
  sentimentDropsBelow: 30,     // Alert if sentiment < 30%
  volumeSpikesAbove: 150,      // Alert if volume > 150% of avg
  angerEmotionExceeds: 40,     // Alert if anger > 40%
  viralMomentumDetected: true  // Alert on viral trends
}

Choose Delivery Channels

Configure where alerts are sent:

Email: Multiple addresses (comma-separated)
Slack: Channel webhooks (#sentiment-alerts)
Webhook: Custom API endpoints

Select Report Types

Choose what to include in alerts:

Executive Summary
Full Analysis
Crisis Alerts Only (recommended for high-frequency monitoring)
Trend Comparison

Alert Threshold Configuration

The Schedule Analysis modal provides granular control over alert conditions:

Sentiment Drop Threshold

<Slider 
  defaultValue={[30]} 
  max={100} 
  step={1} 
/>

Use case: Alert when positive sentiment falls below 30% (indicating majority negative/neutral sentiment).

Volume Spike Detection

<Slider 
  defaultValue={[150]} 
  max={500} 
  step={10} 
/>

Use case: Alert when volume exceeds 150% of the hourly average (viral trend detection).

Emotion-Specific Triggers

<Slider 
  label="Anger emotion exceeds" 
  defaultValue={[40]} 
  max={100} 
  step={1} 
/>

Use case: Trigger crisis alert when anger surpasses 40% (indicates public outrage).

You can combine multiple thresholds. For example: Alert only if both anger > 40% and volume spikes > 150%.

Volatility Score

Volatility measures sentiment stability over time (0-100 scale):

// Source: mockData.ts
{
  volatility: 91,  // Extremely volatile
  crisisLevel: 'high'
}

Interpreting Volatility

0-40: Stable sentiment, predictable trends
41-70: Moderate fluctuations, evolving narrative
71-100: Highly volatile, rapid sentiment shifts

Crisis correlation: Topics with volatility > 70 have a 3x higher chance of becoming crises.

The volatility chart displays 20 mini-bars showing recent fluctuations. Height = sentiment change magnitude.

Crisis Response Workflow

When a crisis is detected:

Verify the Alert

Check the Crisis & Alerts Log for the trigger:

What emotion spiked? (anger, fear, disgust)
What’s the volume change?
What are people saying? (read Top Voices / Pointers)

Analyze Root Cause

Use the AI Insights panel to generate a strategic analysis:

Click “Generate Report” to get LLM-powered recommendations
Review “Key Concerns Identified” and “Risk Factors to Monitor”

Monitor Top Phrases

Review the Top Voices panel for specific complaints:

Negative pointers show recurring criticisms
Positive pointers reveal what’s working

Example: “‘Poor handling of the situation’ (84 mentions)”

Track Live Feed

Read raw posts in the Recent Activity panel to understand tone and context.

Take Action

Based on AI Insights recommendations:

Engage proactively on social media
Issue official statements
Address specific concerns (pricing, policy, transparency)
Monitor sentiment shift after response

Integration with Other Features

Crisis → AI Insights

When a crisis is detected, the AI Insights panel automatically includes crisis-specific guidance:

## Recommended Actions
**Monitor and document** — track the anger signals in real-time
**Engage proactively** — address leading concerns with transparent communication
**Establish rapid response protocols** — Immediate crisis communication is recommended.

Crisis → Export Reports

Export crisis data for stakeholder briefings:

// CSV Export includes:
{
  title: 'Global Food Prices',
  sentiment: 'negative',
  crisisLevel: 'high',
  volatility: 91,
  summary: 'Outrage at corporate profiteering...'
}

Best Practices

Set Conservative Thresholds

Start with sensitive thresholds (e.g., sentiment < 40%, anger > 30%) to catch early warnings. You can relax them if you get too many false positives.

Monitor High-Stakes Topics Daily

For brand launches, policy announcements, or PR-sensitive topics, set hourly monitoring with Crisis Alerts Only enabled.

Cross-Check Emotion + Volume

A 50% anger spike with low volume (< 1K mentions) may not warrant crisis response. Always consider scale.

Use Webhooks for Automation

Integrate with PagerDuty, Opsgenie, or custom incident management systems for 24/7 crisis coverage.

Limitations

Keyword-based detection: The current system uses keyword matching, which can miss sarcasm, context, or subtle negativity. For higher accuracy, consider integrating transformer-based models (BERT, RoBERTa) for emotion classification.No historical baseline: Crisis levels are absolute, not relative to the topic’s normal sentiment. A consistently negative topic (e.g., “Tax Increases”) may always show “medium” crisis even when stable.Platform bias: X/Twitter tends to amplify outrage more than YouTube. Cross-reference multiple platforms before declaring a crisis.

Emotion Classification - Understand which emotions trigger crises
Scheduled Monitoring - Automate crisis detection
AI Insights - Get strategic crisis response recommendations

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

Crisis Detection & Alerts

Overview

Crisis Levels

None

Low

Medium

High

Detection Algorithm

Real-World Crisis Examples

Example 1: Climate Summit (High Crisis)

Example 2: Food Prices (High Crisis)

Example 3: AI Regulation (Medium Crisis)

Crisis & Alerts Log

Negative Spike Detected (🔴 Red)

Sentiment Shift (🟡 Yellow)

Stable / Positive (🔵 Blue)

Scheduled Monitoring & Alerts

Alert Threshold Configuration

Sentiment Drop Threshold

Volume Spike Detection

Emotion-Specific Triggers

Volatility Score

Interpreting Volatility

Crisis Response Workflow

Integration with Other Features

Crisis → AI Insights

Crisis → Export Reports

Best Practices

Set Conservative Thresholds

Monitor High-Stakes Topics Daily

Cross-Check Emotion + Volume

Use Webhooks for Automation

Limitations

Build docs developers (and LLMs) love

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

​Overview

​Crisis Levels

None

Low

Medium

High

​Detection Algorithm

​Real-World Crisis Examples

​Example 1: Climate Summit (High Crisis)

​Example 2: Food Prices (High Crisis)

​Example 3: AI Regulation (Medium Crisis)

​Crisis & Alerts Log

​Negative Spike Detected (🔴 Red)

​Sentiment Shift (🟡 Yellow)

​Stable / Positive (🔵 Blue)

​Scheduled Monitoring & Alerts

​Alert Threshold Configuration

​Sentiment Drop Threshold

​Volume Spike Detection

​Emotion-Specific Triggers

​Volatility Score

​Interpreting Volatility

​Crisis Response Workflow

​Integration with Other Features

​Crisis → AI Insights

​Crisis → Export Reports

​Best Practices

Set Conservative Thresholds

Monitor High-Stakes Topics Daily

Cross-Check Emotion + Volume

Use Webhooks for Automation

​Limitations

​Related Features

Build docs developers (and LLMs) love

Overview

Crisis Levels

Detection Algorithm

Real-World Crisis Examples

Example 1: Climate Summit (High Crisis)

Example 2: Food Prices (High Crisis)

Example 3: AI Regulation (Medium Crisis)

Crisis & Alerts Log

Negative Spike Detected (🔴 Red)

Sentiment Shift (🟡 Yellow)

Stable / Positive (🔵 Blue)

Scheduled Monitoring & Alerts

Alert Threshold Configuration

Sentiment Drop Threshold

Volume Spike Detection

Emotion-Specific Triggers

Volatility Score

Interpreting Volatility

Crisis Response Workflow

Integration with Other Features

Crisis → AI Insights

Crisis → Export Reports

Best Practices

Limitations

Related Features