Lecture 7 - Affective Computing + Personality Traits

What is affective computing?

📖 Term coined by Rosalind Picard in 1994.
💡 Computing that relates to, arises from, or influences emotions.
🤖 Goal: Make machines recognize, interpret, and respond to human emotions.

Key research areas in affective computing

🔍 Challenges:
1⃣ Representation – How to define emotions?
2⃣ Detection – How to identify emotions from users?
3⃣ Classification – How to categorize emotions?
4⃣ Generation – How to make machines express emotions?

Emotions, moods, personality

📌 Time-based classification:
✔ Short-term – Emotions (dominant & specific, e.g., excitement).
✔ Medium-term – Moods (lasting but unnoticed, e.g., feeling cheerful).
✔ Long-term – Personality (stable traits, e.g., introversion).

Emotional expression

✔ Obvious expressions – Facial expressions, voice tone, body posture.
✔ Less obvious expressions – Heart rate, sweat, respiration, pupil dilation.
📌 What we express may not always match what we feel

Emotion recognition system

🛠 Inputs → 🎯 Model → 😊 Outputs
📌 Example system:
✔ Inputs: Video, audio, sensors.
✔ Model: Neural Networks, SVM, etc.
✔ Outputs: Recognized emotions (happy, sad, angry).

Emotion recognition system

🎤 Audio: Speech tone, pitch, pauses, laughter, crying.
🎭 Vision: Facial expressions, gestures, body movement.
🧠 Biosignals: Brain activity, heart rate, muscle tension.

Emotion Classification Methods

📌 Common Machine Learning Techniques:
✔ Hidden Markov Models (HMM) – Tracks sequences over time.
✔ Long Short-Term Memory (LSTM) – Learns from past data.
✔ Convolutional Neural Networks (CNN) – Good for vision-based detection.
✔ Support Vector Machines (SVM) – Finds patterns in data.
✔ k-Nearest Neighbors (KNN) – Finds similar past cases.

Common emotion models

📌 Ways to categorize emotions in computing:
✔ Label-based models – Uses predefined categories (e.g., "happy," "sad").
✔ Ekman Model – 6 universal emotions (joy, anger, sadness, surprise, disgust, fear).
✔ Circumplex Model – Maps emotions in a circular space (valence & arousal).
✔ Plutchik’s Wheel – 8 primary emotions with intensity variations.
✔ PAD/VAD Model – Uses Pleasure, Arousal, and Dominance to classify emotions.

Emotion synthesis + UI design

🎭 Can computers express emotions?
📌 Multimodal outputs (talking avatars, voice tone, gestures).
📌 Challenges:
✔ Uncanny Valley: Too realistic = creepy.
✔ Non-anthropomorphic design: Avoid making computers act too human-like.

Rational choice theory in decision-making

🤖 Assumption: People make logical decisions based on expected value.
📌 Formula:
🔹 U = p(x1)U(x1) + p(x2)U(x2) + … + p(xn)U(xn)
❌ Problem: Humans don’t always make rational choices due to emotions.

Decision affect theory

🎯 People choose based on expected pleasure, not just logical utility!
✔ Rewards influence future decisions.
✔ Surprise, disappointment, and regret play roles in decision-making.

📌 Example: People may prefer a low-risk, low-reward choice if the high-risk one caused stress before.

How affective computing helps users?

📌 Use Cases:
✔ Personalized User Interfaces – Adaptive content based on mood/personality.
✔ Healthcare – Emotion tracking in mental health applications.
✔ Customer Service Bots – AI agents that respond to users’ emotions.
✔ Gaming & VR – Immersive experiences that react to user emotions.

📌 Developed by Paul Ekman.
✔ Identifies 6 universal emotions across all cultures:
😊 Happiness
😠 Anger
😢 Sadness
😲 Surprise
😨 Fear
🤢 Disgust

✅ Strength: Recognized worldwide.
❌ Weakness: Ignores complex emotions like jealousy & love.

Circumplex Model of Affect

📌 Developed by James Russell.
✔ Emotions are mapped in a circular space based on:

• Valence (pleasant → unpleasant)

• Arousal (high energy → low energy)

📌 Examples:
😊 Happy (High Arousal, Positive Valence)
😴 Calm (Low Arousal, Positive Valence)
😡 Angry (High Arousal, Negative Valence)
😭 Sad (Low Arousal, Negative Valence)

✅ Strength: Shows relationships between emotions.
❌ Weakness: Doesn’t differentiate between emotions with the same valence/arousal.

Plutchik’s Wheel of Emotions

📌 Developed by Robert Plutchik.
✔ 8 primary emotions arranged in pairs of opposites:
❤ Love (Joy + Trust)
😂 Optimism (Joy + Anticipation)
😡 Aggressiveness (Anger + Anticipation)
😨 Terror (Fear + Surprise)
😞 Disappointment (Sadness + Surprise)
🤮 Disgust (Sadness + Disgust)

📌 Key Feature: Emotions vary in intensity (e.g., annoyance → anger → rage).

✅ Strength: Shows emotion intensity & blending.
❌ Weakness: Doesn’t explain emotions influenced by culture.

PAD/VAD Emotion Model

📌 PAD: Pleasure, Arousal, Dominance
📌 VAD: Valence, Arousal, Dominance

✔ Used to measure emotions continuously, rather than categorically.
✔ Helps in sentiment analysis & AI models.

📌 Example:
😊 Happy: High Pleasure, High Arousal, Medium Dominance.
😡 Angry: Low Pleasure, High Arousal, High Dominance.
😭 Sad: Low Pleasure, Low Arousal, Low Dominance.

✅ Strength: Works well for machine learning models.
❌ Weakness: Hard to measure accurately in real-world scenarios.

PANA Mood Model

📌 Divides affect into:
✔ Positive Affect (PA): Energy, enthusiasm, and alertness.
✔ Negative Affect (NA): Distress, fear, and nervousness.

📌 4 Categories:
1⃣ PAA (Positive Affect Activated): Active, enthusiastic.
2⃣ PAD (Positive Affect Deactivated): Calm, relaxed.
3⃣ NAA (Negative Affect Activated): Angry, tense.
4⃣ NAD (Negative Affect Deactivated): Depressed, bored.

✅ Strength: Captures emotional changes over time.
❌ Weakness: Doesn't explain complex emotional states.

Big Five Personality Model

📌 Used in psychology to define personality traits.
✔ OCEAN:
1⃣ Openness to Experience – Curious, artistic.
2⃣ Conscientiousness – Organized, responsible.
3⃣ Extraversion – Outgoing, social.
4⃣ Agreeableness – Kind, cooperative.
5⃣ Neuroticism – Anxious, emotionally unstable.

✅ Strength: Explains long-term emotional tendencies.
❌ Weakness: Doesn’t explain situational emotions.