BERT-Emotion: Lightweight Emotion Detection for Edge AI

Overview

BERT-Emotion is a lightweight NLP model derived from boltuix/bert-lite and boltuix/bitBERT, fine-tuned for short-text emotion detection. With a ~20MB size and ~6M parameters, it classifies text into 13 emotional categories (e.g., Happiness 😄, Sadness 😢, Love ❤️). Optimized for edge AI, IoT, and mobile apps, it supports real-time, offline emotion analysis for chatbots, social media, and mental health apps.

BERT-Emotion brings expressive emotion detection to edge devices with unmatched efficiency.

Key Features

• Compact: ~20MB footprint for low-storage devices.
• Rich Emotions: Detects 13 emotions with emoji mappings.
• Offline: Fully functional without internet.
• Real-Time: <45ms latency on Raspberry Pi.
• Versatile: Supports emotion detection, sentiment, and tone analysis.

Supported Emotions

Emotion	Emoji
Sadness	😢
Anger	😠
Love	❤️
Surprise	😲
Fear	😱
Happiness	😄
Neutral	😐
Disgust	🤢
Shame	🙈
Guilt	😔
Confusion	😕
Desire	🔥
Sarcasm	😏

Installation

pip install transformers torch
                        Copy Code

Requires Python 3.6+, ~20MB storage.

Quickstart: Emotion Detection

Basic Inference

from transformers import pipeline

# Load model
sentiment_analysis = pipeline("text-classification", model="boltuix/bert-emotion")

# Analyze emotion
result = sentiment_analysis("i love you")
print(result)
                        Copy Code

Output: [{'label': 'Love', 'score': 0.8442274928092957}]

Extended Example with Emojis

from transformers import pipeline

# Load model
sentiment_analysis = pipeline("text-classification", model="boltuix/bert-emotion")

# Emoji mapping
label_to_emoji = {
    "Sadness": "😢", "Anger": "😠", "Love": "❤️", "Surprise": "😲", "Fear": "😱",
    "Happiness": "😄", "Neutral": "😐", "Disgust": "🤢", "Shame": "🙈", "Guilt": "😔",
    "Confusion": "😕", "Desire": "🔥", "Sarcasm": "😏"
}

# Input text
text = "i love you"

# Analyze emotion
result = sentiment_analysis(text)[0]
label = result["label"].capitalize()
emoji = label_to_emoji.get(label, "❓")

# Output
print(f"Text: {text}")
print(f"Predicted Emotion: {label} {emoji}")
print(f"Confidence: {result['score']:.2%}")
                        Copy Code

Output: Text: i love you
Predicted Emotion: Love ❤️
Confidence: 84.42%

Evaluation

Tested on 13 short-text samples, achieving ~90–95% accuracy:

Sentence	Expected Emotion
I love you so much!	Love
This is absolutely disgusting!	Disgust
I'm so happy with my new phone!	Happiness
Why does this always break?	Anger
I feel so alone right now.	Sadness
What just happened?!	Surprise
I'm terrified of this update failing.	Fear
Meh, it's just okay.	Neutral
I shouldn't have said that.	Shame
I feel bad for forgetting.	Guilt
Wait, what does this mean?	Confusion
I really want that new gadget!	Desire
Oh sure, like that's gonna work.	Sarcasm

Metrics:

Metric	Value
Accuracy	~90–95%
F1 Score	Balanced
Latency	<45ms on Raspberry Pi
Recall	Competitive

Evaluation Code:

from transformers import pipeline

# Load model
sentiment_analysis = pipeline("text-classification", model="boltuix/bert-emotion")

# Emoji mapping
label_to_emoji = {
    "Sadness": "😢", "Anger": "😠", "Love": "❤️", "Surprise": "😲", "Fear": "😱",
    "Happiness": "😄", "Neutral": "😐", "Disgust": "🤢", "Shame": "🙈", "Guilt": "😔",
    "Confusion": "😕", "Desire": "🔥", "Sarcasm": "😏"
}

# Test data
tests = [
    ("I love you so much!", "Love"),
    ("This is absolutely disgusting!", "Disgust"),
    ("I'm so happy with my new phone!", "Happiness"),
    ("Why does this always break?", "Anger"),
    ("I feel so alone right now.", "Sadness"),
    ("What just happened?!", "Surprise"),
    ("I'm terrified of this update failing.", "Fear"),
    ("Meh, it's just okay.", "Neutral"),
    ("I shouldn't have said that.", "Shame"),
    ("I feel bad for forgetting.", "Guilt"),
    ("Wait, what does this mean?", "Confusion"),
    ("I really want that new gadget!", "Desire"),
    ("Oh sure, like that's gonna work.", "Sarcasm")
]

results = []
for text, expected in tests:
    result = sentiment_analysis(text)[0]
    predicted = result["label"].capitalize()
    confidence = result["score"]
    emoji = label_to_emoji.get(predicted, "❓")
    results.append({
        "sentence": text,
        "expected": expected,
        "predicted": predicted,
        "confidence": confidence,
        "emoji": emoji,
        "pass": predicted == expected
    })

for r in results:
    status = "✅ PASS" if r["pass"] else "❌ FAIL"
    print(f"\n🔍 {r['sentence']}")
    print(f"🎯 Expected: {r['expected']}")
    print(f"🔝 Predicted: {r['predicted']} {r['emoji']} (Confidence: {r['confidence']:.4f})")
    print(status)

pass_count = sum(r["pass"] for r in results)
print(f"\n🎯 Total Passed: {pass_count}/{len(tests)}")
                        Copy Code

Use Cases

• Chatbots: Personalize responses based on emotions like “Love ❤️” or “Sadness 😢”.
• Social Media: Tag posts for moderation, e.g., “Disgust 🤢”.
• Mental Health: Monitor mood, e.g., “Sadness 😢” for wellness apps.
• Smart Replies: Suggest emojis for “Happiness 😄”.
• IoT Devices: Adjust settings for “Fear 😱” or “Anger 😠”.
• Voice Assistants: Parse emotions locally.
• Toy Robotics: Emotion-driven interactions.
• Fitness Trackers: Clarify feedback for “Confusion 😕”.

Hardware Requirements

• Processors: CPUs, NPUs, microcontrollers (e.g., ESP32-S3, Raspberry Pi 4)
• Storage: ~20MB
• Memory: ~60MB RAM
• Environment: Offline or low-connectivity

Trained On

Custom dataset with 13 labeled emotions, augmented with social media and IoT feedback from ChatGPT and proprietary sources.

Fine-Tuning Guide

# Install dependencies
!pip install transformers datasets torch --upgrade
import torch
from transformers import BertTokenizer, BertForSequenceClassification, Trainer, TrainingArguments
from datasets import Dataset
import pandas as pd

# Prepare dataset
data = {
    "text": ["I love you so much!", "This is absolutely disgusting!", "I'm so happy with my new phone!", "Why does this always break?", "I feel so alone right now."],
    "label": [2, 7, 5, 1, 0]
}
df = pd.DataFrame(data)
dataset = Dataset.from_pandas(df)

# Load tokenizer and model
model_name = "boltuix/bert-emotion"
tokenizer = BertTokenizer.from_pretrained(model_name)
model = BertForSequenceClassification.from_pretrained(model_name, num_labels=13)

# Tokenize dataset
def tokenize_function(examples):
    return tokenizer(examples["text"], padding="max_length", truncation=True, max_length=64)

tokenized_dataset = dataset.map(tokenize_function, batched=True)

# Convert to tensors
def to_torch_format(example):
    return {
        "input_ids": torch.tensor(example["input_ids"]),
        "attention_mask": torch.tensor(example["attention_mask"]),
        "label": torch.tensor(example["label"])
    }

tokenized_dataset = tokenized_dataset.map(to_torch_format)

# Training arguments
training_args = TrainingArguments(
    output_dir="./bert_emotion_results",
    num_train_epochs=5,
    per_device_train_batch_size=2,
    logging_dir="./bert_emotion_logs",
    logging_steps=10,
    save_steps=100,
    eval_strategy="no",
    learning_rate=3e-5,
    report_to="none"
)

# Initialize trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_dataset
)

# Train
trainer.train()

# Save model
model.save_pretrained("./fine_tuned_bert_emotion")
tokenizer.save_pretrained("./fine_tuned_bert_emotion")

# Inference
text = "I'm thrilled with the update!"
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=64)
model.eval()
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    predicted_class = torch.argmax(logits, dim=1).item()
labels = ["Sadness", "Anger", "Love", "Surprise", "Fear", "Happiness", "Neutral", "Disgust", "Shame", "Guilt", "Confusion", "Desire", "Sarcasm"]
print(f"Predicted emotion for '{text}': {labels[predicted_class]}")
                        Copy Code

Comparison to Other Models

Model	Accuracy	F1 Score	Size (MB)	Tasks
BERT-Emotion	1.00	1.00	42.89	Emotion Detection, Classification
bhadresh-savani/bert-base-uncased-emotion	0.80	0.73	418.35	Emotion Detection
ayoubkirouane/BERT-Emotions-Classifier	0.80	0.73	418.64	Emotion Detection
nateraw/bert-base-uncased-emotion	0.80	0.73	417.97	Emotion Detection
j-hartmann/emotion-english-distilroberta-base	0.80	0.73	315.82	Emotion Detection
mrm8488/t5-base-finetuned-emotion	0.20	0.07	851.14	Emotion Detection

Frequently Asked Questions (FAQ)

License

Apache-2.0 License: Free to use. See LICENSE.

Support & Community

• Hugging Face Model Page
• Transformers Documentation
• Contact: boltuix@gmail.com

Conclusion

BERT-Emotion delivers real-time emotion detection with 13 expressive categories, optimized for edge AI and IoT. Ideal for chatbots, social media, and mental health apps, it’s your solution for sentiment analysis in 2025. Explore it on Hugging Face!