Understanding Metrics

Overview

SENTi-radar uses a combination of keyword-based emotion detection, statistical analysis, and AI-powered summarization to provide real-time sentiment insights. This guide explains the algorithms and formulas behind each metric.

Emotion Detection System

How It Works

SENTi-radar analyzes text from all sources (X posts, Reddit comments, YouTube comments, news headlines) using a keyword lexicon scoring system. The implementation can be found in TopicDetail.tsx:146-167.

The Six Core Emotions

Based on Ekman’s universal emotions framework:

Fear (35% avg in crisis topics)

Keywords: fear, scared, worried, panic, threat, risk, dangerous, crisis, collapse, shortage, anxiety, alarm, uncertainty, instability, warn, catastroph, turmoil, chaos, tension, war, nuclear, invasion, missile, attack, afraid, terrifying, dread, horrified, alarmingColor: Orange (bg-fear)Common in: Geopolitical conflicts, health crises, economic uncertainty, supply chain disruptions

Anger (48% avg in crisis topics)

Keywords: anger, angry, outrage, furious, rage, frustrat, unacceptable, scandal, corrupt, condemn, protest, exploit, injustice, blame, backlash, fury, demand, ban, oppose, ridiculous, pathetic, disgusting, shameful, hate, upset, terrible, horrible, awful, liarColor: Red (bg-anger)Common in: Policy debates, corporate scandals, social justice issues, price inflation topics

Sadness (25% avg in negative topics)

Keywords: sad, disappoint, tragic, loss, suffer, grief, regret, devastat, despair, victim, casualt, death, pain, mourn, unfortunate, heartbreak, sorrow, crying, tears, sorry, depressing, hopelessColor: Blue (bg-sadness)Common in: Climate discussions, humanitarian crises, personal finance struggles, loss events

Joy (45% avg in positive topics)

Keywords: happy, excited, great, amazing, love, excellent, fantastic, celebrate, breakthrough, success, innovation, optimis, hopeful, launch, growth, improve, wonderful, awesome, congratulations, proud, thrilled, wow, incredible, blessed, thank, gladColor: Green (bg-joy)Common in: Product launches, tech breakthroughs, positive news, achievement announcements

Surprise (25% avg in mixed topics)

Keywords: shocking, unexpected, unbelievable, stunning, incredible, reveal, bombshell, breaking, unprecedented, remarkable, wtf, omg, cant believe, seriously, really, whoa, wait whatColor: Yellow (bg-surprise)Common in: Breaking news, unexpected announcements, viral moments, plot twists

Disgust (10% avg in scandal topics)

Keywords: disgust, appalling, horrible, corrupt, toxic, vile, sickening, revolting, gross, nauseating, shameful, pathetic, ridiculuousColor: Purple (bg-disgust)Common in: Corruption scandals, unethical behavior, food safety issues, abuse revelations

Scoring Algorithm

function scoreEmotions(texts: string[]): EmotionData[] {
  const allText = texts.join(' ').toLowerCase();
  const scores: Record<string, number> = {};
  
  // Count keyword matches for each emotion
  for (const [emotion, words] of Object.entries(EMOTION_KEYWORDS)) {
    scores[emotion] = words.reduce((sum, keyword) => {
      const regex = new RegExp(keyword, 'gi');
      return sum + (allText.match(regex)?.length || 0);
    }, 0);
  }
  
  // Convert to percentages
  const total = Object.values(scores).reduce((a, b) => a + b, 0) || 1;
  const emotions = Object.entries(scores)
    .map(([emotion, score]) => ({
      emotion,
      percentage: Math.round((score / total) * 100),
      count: score,
    }))
    .sort((a, b) => b.percentage - a.percentage);
  
  // Normalize to exactly 100%
  const sum = emotions.reduce((s, e) => s + e.percentage, 0);
  if (sum !== 100 && sum > 0) emotions[0].percentage += (100 - sum);
  
  return emotions;
}

Key Points:

Case-insensitive matching
Partial word matches (e.g., “frustrat” matches “frustrated”, “frustrating”)
Multiple occurrences count multiple times
Results always sum to exactly 100%
Emotions sorted by dominance (highest percentage first)

The dominant emotion is always emotions[0] and the second emotion is emotions[1] after sorting.

Sentiment Classification

Three Sentiment States

SENTi-radar classifies overall sentiment into three categories:

Positive
Negative
Mixed

Criteria: Positive keyword count > negative keyword count × 1.3Positive Keywords (from TopicDetail.tsx:264):

launch, success, growth, celebrate, innovation
deal, partnership, breakthrough, improve
great, amazing, wonderful, fantastic

Typical Emotion Profile:

Joy: 40-50%
Surprise: 20-30%
Low anger/fear/sadness

Example Topics: Product launches, tech breakthroughs, positive policy announcements

Criteria: Negative keyword count > positive keyword count × 1.3Negative Keywords (from TopicDetail.tsx:263):

war, attack, crisis, shortage, tension, conflict
scandal, ban, protest, threat, crash, decline
fail, corrupt, dangerous

Typical Emotion Profile:

Anger: 40-50%
Fear: 20-35%
Sadness: 15-25%

Example Topics: Geopolitical conflicts, economic crises, corporate scandals, climate issues

Sentiment Calculation Code

const negKw = ['war', 'attack', 'crisis', 'shortage', 'tension', 'conflict', ...];
const posKw = ['launch', 'success', 'growth', 'celebrate', 'innovation', ...];

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

const sentiment = 
  negCount > posCount * 1.3 ? 'negative' : 
  posCount > negCount * 1.3 ? 'positive' : 
  'mixed';

The 1.3x multiplier prevents minor differences from triggering a sentiment classification. A clear dominance is required.

Crisis Level Detection

SENTi-radar automatically detects crisis situations based on negative keyword frequency.

The Four Crisis Levels

Level	Criteria	Negative Keyword Count	Visual Indicator	Typical Response
None	Low concern	0-1 keywords	🔵 Blue	Monitor normally
Medium	Elevated concern	2-3 keywords	🟡 Yellow	Watch closely
High	Crisis risk	4+ keywords	🔴 Red	Immediate attention

Crisis Detection Code

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

Crisis Alerts

When crisis level is medium or high, SENTi-radar displays real-time alerts: High Crisis Alerts:

🔴 Negative spike detected
Pulsing red indicator
Shows dominant negative emotion
Timestamp: “Just now”

Medium Crisis Alerts:

🟡 Sentiment Shift
Amber indicator
“Sudden increase in conversation volume”
Timestamp: “30 min ago”

Volatility Score

What It Measures

Volatility indicates how rapidly conversation volume is changing. High volatility suggests:

Breaking news or viral moments
Rapid sentiment shifts
Potential crisis escalation
High public attention

Score Range

0-40: Low volatility (stable conversation)
41-70: Moderate volatility (growing interest)
71-100: High volatility (rapid acceleration, potential crisis)

Visual Representation

The volatility widget shows:

20-bar sparkline chart with random heights (simulating volume spikes)
Risk label: “High” (volatility >70) or “Moderate” (≤70)
Numerical score: “Volatility at X/100”

Topics with volatility >80 should be refreshed frequently (every 5-10 minutes) during live monitoring.

Volume & Change Metrics

Total Mentions Volume

Displayed in abbreviated format:

function formatVolume(n: number): string {
  if (n >= 1_000_000) return (n / 1_000_000).toFixed(1) + 'M';
  if (n >= 1_000) return (n / 1_000).toFixed(1) + 'K';
  return n.toString();
}

Examples:

284500 → 284.5K
1200000 → 1.2M
531000 → 531.0K

Hourly Change Percentage

Shows conversation growth/decline in the last hour:

Positive change (e.g., +42%): 🔼 Green, trending up arrow
Negative change (e.g., -15%): 🔽 Red, trending down arrow

Change percentage is calculated against the previous 1-hour window, not the absolute baseline.

Theme Detection

SENTi-radar automatically categorizes topics into 7 thematic areas:

The Seven Themes

Geopolitical

Keywords: war, tension, conflict, iran, israel, russia, ukraine, china, nato, missile, nuclear, sanction, military, attack, defense, border, invasion, ceasefire, diplomacy, treaty, army, troopsKey Takeaway Templates:

Escalation fears are driving market volatility
Diplomatic channels remain under pressure
Defense and security discussions dominate
Economic ripple effects are a major worry
International community response is being closely watched

Energy

Keywords: oil, gas, fuel, energy, opec, crude, petroleum, shortage, reserve, pipeline, refinery, barrel, lng, solar, renewable, lpg, petrol, dieselKey Takeaway Templates:

Fuel price hikes are the #1 concern
Energy security is being questioned
Calls for strategic reserve deployment are intensifying
Industry impact is significant
Government policy response is under heavy public scrutiny

Policy

Keywords: guideline, regulation, law, policy, government, ministry, ugc, nep, bill, act, reform, amendment, mandate, compliance, rule, directive, framework, education, university, collegeKey Takeaway Templates:

Stakeholders are divided
Implementation timeline and enforcement are major debate points
Affected communities are mobilizing
Legal challenges and constitutional questions are being raised
Comparative analysis with international standards shows gaps and improvements

Tech

Keywords: phone, galaxy, iphone, laptop, app, software, ai, robot, chip, launch, release, samsung, apple, google, tesla, nvidia, startup, feature, update, device, processor, gpuKey Takeaway Templates:

Early adopters are buzzing
Comparisons with competitors are driving heated debates
Innovation claims are being scrutinized
Pricing strategy is polarizing
Supply chain and availability concerns are building

Economic

Keywords: market, stock, inflation, recession, economy, gdp, trade, tariff, unemployment, interest, rate, bank, fiscal, budget, debt, investment, growth, crash, rupee, dollarKey Takeaway Templates:

Market sentiment is fragile
Inflation concerns are hitting household budgets
Central bank policy decisions are being analyzed
Employment outlook is uncertain
Global trade dynamics are shifting

Health

Keywords: health, covid, vaccine, hospital, disease, epidemic, pandemic, medical, drug, treatment, doctor, patient, mental, who, outbreak, virusKey Takeaway Templates:

Public health preparedness is being questioned
Healthcare accessibility is a key concern
Trust in health institutions is being tested
Preventive measures and personal health awareness are trending
Policy response speed and transparency are under intense scrutiny

Social

Theme Selection Algorithm

let bestTheme = 'general';
let bestScore = 0;

for (const [theme, config] of Object.entries(TOPIC_THEMES)) {
  const score = config.keywords.filter(kw => text.includes(kw)).length;
  if (score > bestScore) {
    bestScore = score;
    bestTheme = theme;
  }
}

The theme with the most keyword matches wins. Takeaways are pulled from the winning theme’s template library.

Data Source Attribution

SENTi-radar displays which sources contributed to the analysis:

Source Combination	Display Label
X + Reddit + YouTube + News	`X · Reddit · YouTube · News`
X + Reddit	`X · Reddit`
X + News	`X · News`
Reddit + News	`Reddit · News`
YouTube + News	`YouTube · News`
News only	`News`
None (fallback)	`Keyword`

The most comprehensive analysis comes from X · Reddit · YouTube · News sources. Configure all API keys for best results.

Confidence & Accuracy

Factors Affecting Accuracy

Sample Size: More texts = more accurate emotion distribution
- 10-30 texts: Low confidence
- 31-100 texts: Medium confidence
- 100+ texts: High confidence
Source Diversity: Multi-source data reduces platform bias
- Single source: Platform-specific echo chamber
- 2-3 sources: Balanced view
- 4 sources: Comprehensive sentiment picture
Keyword Coverage: More emotion keywords matched = higher confidence
- 0-5 keywords: Likely neutral/unclear
- 6-15 keywords: Clear sentiment pattern
- 16+ keywords: Very strong sentiment signal

When to Trust Metrics

✅ High Trust Scenarios:

100+ texts analyzed from 3+ sources
Clear emotion dominance (>40% for top emotion)
Multiple crisis keywords detected
Sentiment and emotion alignment (e.g., negative sentiment + anger/fear dominant)

⚠️ Lower Trust Scenarios:

<30 texts analyzed
Single source only (X or Reddit unavailable)
Flat emotion distribution (all <20%)
Sentiment/emotion mismatch (e.g., positive sentiment but anger dominant)

Analyzing Topics - Step-by-step guide to running sentiment analysis
Exporting Data - Export metrics data for external analysis
Adding Data Sources - Configure API keys to improve accuracy

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

Understanding Metrics

Overview

Emotion Detection System

How It Works

The Six Core Emotions

Scoring Algorithm

Sentiment Classification

Three Sentiment States

Sentiment Calculation Code

Crisis Level Detection

The Four Crisis Levels

Crisis Detection Code

Crisis Alerts

Volatility Score

What It Measures

Score Range

Visual Representation

Volume & Change Metrics

Total Mentions Volume

Hourly Change Percentage

Theme Detection

The Seven Themes

Theme Selection Algorithm

Data Source Attribution

Confidence & Accuracy

Factors Affecting Accuracy

When to Trust Metrics

Build docs developers (and LLMs) love

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

​Overview

​Emotion Detection System

​How It Works

​The Six Core Emotions

​Scoring Algorithm

​Sentiment Classification

​Three Sentiment States

​Sentiment Calculation Code

​Crisis Level Detection

​The Four Crisis Levels

​Crisis Detection Code

​Crisis Alerts

​Volatility Score

​What It Measures

​Score Range

​Visual Representation

​Volume & Change Metrics

​Total Mentions Volume

​Hourly Change Percentage

​Theme Detection

​The Seven Themes

​Theme Selection Algorithm

​Data Source Attribution

​Confidence & Accuracy

​Factors Affecting Accuracy

​When to Trust Metrics

​Related Resources

Build docs developers (and LLMs) love

Overview

Emotion Detection System

How It Works

The Six Core Emotions

Scoring Algorithm

Sentiment Classification

Three Sentiment States

Sentiment Calculation Code

Crisis Level Detection

The Four Crisis Levels

Crisis Detection Code

Crisis Alerts

Volatility Score

What It Measures

Score Range

Visual Representation

Volume & Change Metrics

Total Mentions Volume

Hourly Change Percentage

Theme Detection

The Seven Themes

Theme Selection Algorithm

Data Source Attribution

Confidence & Accuracy

Factors Affecting Accuracy

When to Trust Metrics

Related Resources