Skip to content

Emotion Detection

Detect specific emotions beyond positive/negative - 8 emotion categories with intensity scoring.

Why emotion detection?

Customer experience analysis, mental health monitoring, social media sentiment deep-dives, and chatbot emotional intelligence. Knowing something is "negative" isn't enough - is the customer angry or sad? Is the tweet fearful or disgusted?

malaysian-manglish-nlp detects 8 emotion categories with intensity, handling Manglish and code-switched text.

Load module

import malaysian_manglish_nlp as mnlp

result = mnlp.detect_emotion("Geram betul aku dengan dia, dah la lambat pastu buat hal")
print(result)
# {'primary': 'anger', 'score': 0.88, 'secondary': 'disgust'}

Basic usage

Single emotion detection

mnlp.detect_emotion("Happy gila dapat result exam!")
# {'primary': 'joy', 'score': 0.91, 'secondary': 'surprise'}

mnlp.detect_emotion("Sedih betul, kucing aku hilang dah 3 hari")
# {'primary': 'sadness', 'score': 0.89, 'secondary': 'fear'}

mnlp.detect_emotion("Seram gila movie tu, tak boleh tidur malam")
# {'primary': 'fear', 'score': 0.85, 'secondary': 'surprise'}

mnlp.detect_emotion("Eww geli nya, makanan ada ulat")
# {'primary': 'disgust', 'score': 0.92, 'secondary': 'anger'}

All 8 emotions

# Joy
mnlp.detect_emotion("Alhamdulillah dapat kerja baru!")
# {'primary': 'joy', 'score': 0.93, 'secondary': 'anticipation'}

# Sadness
mnlp.detect_emotion("Rindu sangat kat arwah mak")
# {'primary': 'sadness', 'score': 0.91, 'secondary': 'trust'}

# Anger
mnlp.detect_emotion("Bodoh punya driver, potong queue!")
# {'primary': 'anger', 'score': 0.90, 'secondary': 'disgust'}

# Fear
mnlp.detect_emotion("Takut nak balik sorang malam ni")
# {'primary': 'fear', 'score': 0.87, 'secondary': 'sadness'}

# Surprise
mnlp.detect_emotion("Eh tak sangka pulak dia datang!")
# {'primary': 'surprise', 'score': 0.88, 'secondary': 'joy'}

# Disgust
mnlp.detect_emotion("Meluat betul tengok perangai dia")
# {'primary': 'disgust', 'score': 0.89, 'secondary': 'anger'}

# Trust
mnlp.detect_emotion("Aku percaya kau boleh buat, you got this!")
# {'primary': 'trust', 'score': 0.84, 'secondary': 'anticipation'}

# Anticipation
mnlp.detect_emotion("Tak sabar nak tunggu Hari Raya ni!")
# {'primary': 'anticipation', 'score': 0.90, 'secondary': 'joy'}

Batch processing

Process multiple texts efficiently:

texts = [
    "Best gila dapat hadiah birthday!",
    "Kecewa betul dengan service dia",
    "Scared gila tengok accident depan mata",
    "Tak sabar nak pergi holiday!",
]

results = mnlp.detect_emotions_batch(texts)
for text, emotion in zip(texts, results):
    print(f"{text[:35]:35s} → {emotion['primary']} ({emotion['score']:.2f})")

# Best gila dapat hadiah birthday!    → joy (0.91)
# Kecewa betul dengan service dia     → sadness (0.82)
# Scared gila tengok accident...      → fear (0.86)
# Tak sabar nak pergi holiday!        → anticipation (0.88)

Emotion summary

Get an overview of emotions across a collection of texts:

reviews = [
    "Sedap gila! Confirm repeat!",
    "Service slow nak mampus",
    "Ok je, nothing special",
    "Best! Will come again for sure",
    "Mahal sangat, rasa biasa je",
]

summary = mnlp.emotion_summary(reviews)
print(summary)
# {'dominant': 'joy',
#  'distribution': {'joy': 2, 'anger': 1, 'neutral': 1, 'disgust': 1},
#  'average_intensity': 0.78}

Social media text examples

# Twitter/X rage
mnlp.detect_emotion("Geram gila dengan J&T, parcel hilang lagi sekali!!!")
# {'primary': 'anger', 'score': 0.93, 'secondary': 'disgust'}

# WhatsApp sadness
mnlp.detect_emotion("Takpelah, aku biasa dah kena macam ni sorang2")
# {'primary': 'sadness', 'score': 0.84, 'secondary': 'trust'}

# Instagram excitement
mnlp.detect_emotion("OMG!!! Finally sampai juga parcel tu! Cantik gila 😍")
# {'primary': 'joy', 'score': 0.92, 'secondary': 'surprise'}

# Reddit frustration
mnlp.detect_emotion("LRT delay lagi, hari2 macam ni, what's the point of paying")
# {'primary': 'anger', 'score': 0.81, 'secondary': 'sadness'}

CLI usage

Emotion detection is primarily a Python API. For CLI:

# Full analysis (includes emotion)
$ mnlp analyze "Geram betul dengan dia"

# With JSON output
$ mnlp analyze "Happy gila!" --json

How it works

  1. Lexicon matching - emotion word dictionaries for BM and EN (3,000+ terms)
  2. Intensifier handling - "gila", "sangat", "betul" boost scores
  3. Context analysis - surrounding words modify emotion classification
  4. Secondary emotion - mixed emotions detected (e.g., anger + disgust)
  5. Code-switching support - BM and EN words processed together

Performance

Metric Score
8-class accuracy 78.4%
Macro F1 74.2%
Joy F1 85.1%
Anger F1 82.3%
Sadness F1 76.8%
Throughput 20,000 texts/sec
Latency (single) < 0.5ms

See also