📊 MoodPulse: Processed Data and Embeddings for Emotion Analysis

MoodPulse provides a self-contained dataset repository for use with the AffectiveLens project—an end-to-end NLP pipeline for emotion detection in text. It includes the full processing stack from raw text to final DistilBERT-based sentence embeddings, allowing researchers to bypass time-consuming preprocessing and directly train or benchmark models.

---

🧾 Dataset Description

This dataset builds upon the original GoEmotions dataset by Google Research, which includes 58k carefully curated Reddit comments labeled with 28 fine-grained emotions.

In MoodPulse, these labels are condensed into three mutually exclusive emotion classes:

• Positive
• Neutral
• Negative

The dataset is structured to support every phase of the AffectiveLens pipeline:

• Raw CSVs
• Tokenized data in Hugging Face datasets format
• Precomputed DistilBERT embeddings

This enables full reproduction of results without requiring re-tokenization or embedding computation.

---

🗂️ Dataset Structure

The dataset is organized into logical folders corresponding to different stages of processing:

/
├── data/
│   ├── full\_dataset/
│   │   ├── goemotions\_1.csv
│   │   ├── goemotions\_2.csv
│   │   └── goemotions\_3.csv
│   │
│   ├── processed/
│   │   ├── GoEmotions\_Tokenized\_Train\_Pool/
│   │   └── GoEmotions\_Tokenized\_Test/
│   │
│   └── embeddings/
│       ├── MentalTrain/
│       └── MentalTest/

📁 Folder Descriptions

• data/full_dataset/
  Original GoEmotions CSV files split into parts.

• data/processed/
  Tokenized datasets using Hugging Face datasets format, ready for embedding extraction.

• data/embeddings/
  Final DistilBERT [CLS] token embeddings for the training and test sets. These are saved as Hugging Face datasets and ready for model input.

---

🚀 How to Use

You can load the tokenized data or precomputed embeddings directly using the Hugging Face datasets library.

from datasets import load_dataset

# Define repository ID and folder to load
repo_id = "psyrishi/MoodPulse"
data_folder = "data/embeddings/MentalTrain"  # or "data/embeddings/MentalTest"

# Load the dataset split
train_embeddings = load_dataset(repo_id, data_dir=data_folder, split='train')

print("Sample entry:")
print(train_embeddings[0])

# Access embeddings and labels
embedding_vector = train_embeddings[0]['cls_embedding']
label_vector = train_embeddings[0]['labels']

💡 Tip: You can replace data_dir to load the tokenized datasets instead, if desired.

---

📌 Use Cases

• Train or benchmark emotion classification models using high-quality, preprocessed embeddings.
• Compare performance of traditional ML models vs. transformer-based models.
• Build emotion-aware applications for mental health, customer feedback, or social media monitoring.

---

📚 Citation

This dataset is a processed derivative of the original GoEmotions dataset:

@inproceedings{demszky2020goemotions,
  title={GoEmotions: A Dataset of Fine-Grained Emotions},
  author={Demszky, Dorottya and Movshovitz-Attias, Dana and Ko, Jeongwoo and Cowen, Alan and Nemade, Gaurav and Ravi, Sujith},
  booktitle={Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL)},
  year={2020}
}

If you use MoodPulse in your work, please cite both the original GoEmotions authors and link back to this repository.

---

⚖️ Licensing

• Original data: Provided under the Creative Commons Attribution 4.0 International (CC BY 4.0) license by Google Research.
• Code and processing logic: Provided under the MIT License.

Please refer to the LICENSE file for full details.

---

🙏 Acknowledgments

Special thanks to Google Research for the creation and open release of the GoEmotions dataset, and to the Hugging Face team for providing the open-source tools that made this processing pipeline possible.

---

🔗 Related Projects

• GoEmotions Dataset (Google)
• AffectiveLens — Emotion detection pipeline built on top of this dataset.

---