tanzeelabbas114/emojibert-sentiment-analysis

Emoji-Fused Sentiment Analysis BERT (94% Accuracy)

Official Model from Scientific Reports (2025)
Paper: Sentiment analysis of emoji fused reviews using machine learning and Bert
DOI: https://doi.org/10.1038/s41598-025-92286-0

Authors: Amit Khan, Dipankar Majumdar, Bikromadittya Mondal

This is the official model that achieved 94% accuracy by preserving and enhancing emojis using the copyrighted EMOJIXT dictionary — instead of removing them like most models do.

---

Key Innovation

We replace emojis with sentiment-rich words:

• Happy emoji → happy joyful
• Crying emoji → crying sad
• Heart emoji → love adore
• Thumbs up emoji → thumbs up good

Result: +9% accuracy gain over state-of-the-art baselines.

---

Labels

Label	Sentiment
0	Negative
1	Neutral
2	Positive

---

Usage

from transformers import pipeline

classifier = pipeline("text-classification", model="tanzeelabbas114/emojibert-sentiment-analysis")

print(classifier("flight delayed again crying sad")) 
# → LABEL_0 (Negative)

print(classifier("best airline ever love adore happy joyful"))
# → LABEL_2 (Positive)

---

Performance

• Accuracy: 94%
• Dataset: Twitter US Airline Sentiment (tweet_eval)
• Baseline Improvement: +11% over BERT without emoji handling

---

Citation

@article{khan2025sentiment,
  title={{Sentiment analysis of emoji fused reviews using machine learning and Bert}},
  author={Khan, Amit and Majumdar, Dipankar and Mondal, Bikromadittya},
  journal={Scientific Reports},
  volume={15},
  pages={7538},
  year={2025},
  doi={10.1038/s41598-025-92286-0}
}

---

Reproduced & uploaded by tanzeelabbas114 — making published research openly accessible!