Sentiment Analysis

AI-powered sentiment analysis detects the emotional tone of user messages, enabling tone-matched advertising and appropriate creative selection.

Overview

Sentiment analysis classifies the emotional tone of user interactions into positive, neutral, or negative categories with a confidence score. This enables bidders to match ad tone to user emotional state for better engagement.

Sentiment Object Structure

{
  "sentiment": {
    "value": "positive",
    "score": 0.8
  }
}

Field Reference

Field	Type	Required	Description	Example
value	enum	Yes	"positive", "neutral", or "negative"	"positive"
score	number	Yes	Sentiment score 0.0-1.0	0.8

Sentiment Values

positive

User expresses positive emotions, enthusiasm, or satisfaction.

Indicators:

• Excited language
• Positive words (great, love, amazing)
• Enthusiastic punctuation (!)
• Expressions of satisfaction

Score Range: 0.6 - 1.0

Example:

{
  "value": "positive",
  "score": 0.85
}

Use Cases:

• Upbeat, energetic creatives
• Celebration-themed ads
• Premium offerings
• Aspirational messaging

Example Messages:

• "This sounds amazing!"
• "I love this idea!"
• "Perfect, exactly what I need!"

neutral

User expresses neutral tone without strong emotional signals.

Indicators:

• Factual questions
• Informational requests
• Objective language
• No emotional markers

Score Range: 0.4 - 0.6

Example:

{
  "value": "neutral",
  "score": 0.5
}

Use Cases:

• Informational creatives
• Straightforward messaging
• Feature-focused ads
• Educational content

Example Messages:

• "What are the options?"
• "Tell me about the features"
• "How does this work?"

negative

User expresses negative emotions, frustration, or dissatisfaction.

Indicators:

• Frustrated language
• Negative words (bad, hate, terrible)
• Problem descriptions
• Complaints

Score Range: 0.0 - 0.4

Example:

{
  "value": "negative",
  "score": 0.3
}

Use Cases:

• Solution-oriented creatives
• Empathetic messaging
• Problem-solving ads
• Support services

Example Messages:

• "This is frustrating"
• "I'm having trouble with..."
• "This doesn't work"

Sentiment Score

The sentiment score provides a continuous measure of emotional intensity.

Range: 0.0 to 1.0

Interpretation:

• 0.8 - 1.0: Very positive
• 0.6 - 0.8: Moderately positive
• 0.4 - 0.6: Neutral
• 0.2 - 0.4: Moderately negative
• 0.0 - 0.2: Very negative

Example:

{
  "score": 0.85
}

Complete Examples

Example 1: Positive Sentiment

User excited about vacation planning:

{
  "ext": {
    "aura": {
      "messageSequenceNumber": 3,
      "messageLength": 42,
      "intent": {
        "value": "purchase_inquiry",
        "confidence": 0.92,
        "topics": ["vacation", "beach resort"],
        "signals": ["expressed_excitement", "ready_to_purchase"]
      },
      "sentiment": {
        "value": "positive",
        "score": 0.9
      }
    }
  }
}

Recommended Creative Approach:

• Upbeat, energetic tone
• Celebration imagery
• Exciting language
• Premium positioning

Example 2: Neutral Sentiment

User researching options:

{
  "ext": {
    "aura": {
      "messageSequenceNumber": 2,
      "messageLength": 35,
      "intent": {
        "value": "information_seeking",
        "confidence": 0.88,
        "topics": ["golf equipment", "pricing"],
        "signals": ["comparison_shopping"]
      },
      "sentiment": {
        "value": "neutral",
        "score": 0.5
      }
    }
  }
}

Recommended Creative Approach:

• Informational tone
• Feature highlights
• Comparison charts
• Objective messaging

Example 3: Negative Sentiment

User frustrated with booking issue:

{
  "ext": {
    "aura": {
      "messageSequenceNumber": 4,
      "messageLength": 58,
      "intent": {
        "value": "support_request",
        "confidence": 0.9,
        "topics": ["booking problem", "cancellation"],
        "signals": ["expressed_frustration", "needs_assistance"]
      },
      "sentiment": {
        "value": "negative",
        "score": 0.3
      }
    }
  }
}

Recommended Creative Approach:

• Empathetic tone
• Solution-focused
• Support services
• Reassuring messaging

Publisher Implementation

Sentiment Detection

class SentimentAnalyzer {
  async analyzeSentiment(message) {
    // Use AI/ML model for sentiment classification
    const classification = await this.classifyMessage(message)

    return {
      value: this.getSentimentValue(classification.score),
      score: classification.score,
    }
  }

  getSentimentValue(score) {
    if (score >= 0.6) return 'positive'
    if (score <= 0.4) return 'negative'
    return 'neutral'
  }

  // Simple rule-based fallback
  analyzeSimple(message) {
    const text = message.toLowerCase()
    let score = 0.5 // Start neutral

    // Positive indicators
    const positiveWords = ['great', 'love', 'amazing', 'perfect', 'excellent']
    positiveWords.forEach(word => {
      if (text.includes(word)) score += 0.1
    })

    // Negative indicators
    const negativeWords = ['bad', 'hate', 'terrible', 'frustrated', 'problem']
    negativeWords.forEach(word => {
      if (text.includes(word)) score -= 0.1
    })

    // Clamp to 0-1 range
    score = Math.max(0, Math.min(1, score))

    return {
      value: this.getSentimentValue(score),
      score: score,
    }
  }
}

Usage Example

const analyzer = new SentimentAnalyzer()

// Positive example
const sentiment1 = await analyzer.analyzeSentiment('This looks amazing!')
// { value: 'positive', score: 0.9 }

// Neutral example
const sentiment2 = await analyzer.analyzeSentiment('What are the options?')
// { value: 'neutral', score: 0.5 }

// Negative example
const sentiment3 = await analyzer.analyzeSentiment('This is frustrating')
// { value: 'negative', score: 0.3 }

Bidder Implementation

Sentiment-Based Creative Selection

function selectCreative(sentiment, creatives) {
  // Filter creatives by sentiment match
  const matchingCreatives = creatives.filter(creative => {
    if (sentiment.value === 'positive') {
      return creative.tone === 'upbeat' || creative.tone === 'energetic'
    } else if (sentiment.value === 'negative') {
      return creative.tone === 'empathetic' || creative.tone === 'supportive'
    } else {
      return creative.tone === 'neutral' || creative.tone === 'informational'
    }
  })

  return matchingCreatives[0] || creatives[0]
}

Bid Adjustment

function adjustBidForSentiment(basePrice, sentiment, intent) {
  let modifier = 1.0

  // Positive sentiment + purchase intent = higher value
  if (sentiment.value === 'positive' && intent.value === 'purchase_inquiry') {
    modifier = 1.3
  }

  // Negative sentiment = lower bid or skip
  if (sentiment.value === 'negative') {
    modifier = 0.7
  }

  // Very positive sentiment = premium
  if (sentiment.score > 0.8) {
    modifier *= 1.2
  }

  return basePrice * modifier
}

Tone Matching

function matchTone(sentiment) {
  const toneMap = {
    positive: {
      high: ['celebratory', 'energetic', 'exciting'],
      medium: ['upbeat', 'friendly', 'warm'],
    },
    neutral: {
      high: ['informational', 'professional', 'clear'],
      medium: ['straightforward', 'factual', 'objective'],
    },
    negative: {
      high: ['empathetic', 'supportive', 'understanding'],
      medium: ['solution-focused', 'helpful', 'reassuring'],
    },
  }

  const intensity =
    sentiment.score > 0.7 || sentiment.score < 0.3 ? 'high' : 'medium'
  return toneMap[sentiment.value][intensity]
}

Sentiment + Intent Combinations

High-Value Combinations

Sentiment	Intent	Strategy
Positive	purchase_inquiry	Premium pricing, aspirational ads
Positive	information_seeking	Engaging educational content
Neutral	comparison	Feature comparison, objective
Negative	support_request	Empathetic support services

Example: Positive + Purchase Intent

{
  "intent": {
    "value": "purchase_inquiry",
    "confidence": 0.92,
    "topics": ["luxury resort"],
    "signals": ["ready_to_purchase", "expressed_excitement"]
  },
  "sentiment": {
    "value": "positive",
    "score": 0.9
  }
}

Strategy:

• Show premium offerings
• Use aspirational imagery
• Emphasize exclusivity
• Higher bids

Example: Negative + Support Request

{
  "intent": {
    "value": "support_request",
    "confidence": 0.88,
    "topics": ["booking issue"],
    "signals": ["expressed_frustration", "needs_assistance"]
  },
  "sentiment": {
    "value": "negative",
    "score": 0.3
  }
}

Strategy:

• Show support services
• Use empathetic tone
• Emphasize solutions
• Lower bids or skip

Best Practices

1. Match Creative Tone

Always match creative tone to sentiment:

function shouldShowCreative(creative, sentiment) {
  if (sentiment.value === 'negative' && creative.tone === 'celebratory') {
    return false // Tone mismatch
  }
  return true
}

2. Combine with Intent

Use sentiment + intent together:

function calculateValue(sentiment, intent) {
  const baseValue = getIntentValue(intent)
  const sentimentModifier = getSentimentModifier(sentiment)
  return baseValue * sentimentModifier
}

3. Handle Edge Cases

Be careful with extreme sentiments:

function shouldBid(sentiment) {
  // Skip very negative sentiment
  if (sentiment.score < 0.2) {
    return false
  }
  return true
}

4. Use Score Granularity

Use the score for fine-grained decisions:

function getCreativeIntensity(sentiment) {
  if (sentiment.score > 0.8) return 'high'
  if (sentiment.score > 0.6) return 'medium'
  if (sentiment.score > 0.4) return 'low'
  if (sentiment.score > 0.2) return 'medium'
  return 'high' // Negative
}

Privacy Considerations

Sentiment analysis is privacy-safe:

• No PII: Only analyzes message content
• Contextual: Based on current conversation
• Aggregated: No long-term tracking
• Transparent: Users understand context is analyzed

Next Steps

• Intent Analysis - User intent signals
• Conversation Metadata - Operational metrics
• Field Reference - Complete field documentation