Pradeep-mahato/ModernBERT-GoEmotions

ModernBERT-GoEmotions

Model Summary

ModernBERT-GoEmotions is a fine-tuned Transformer-based model for multi-label emotion classification. Given a short text input, the model predicts one or more emotions from a predefined set of 27 emotions + neutral, as defined in the GoEmotions dataset.

The model is designed for applications requiring fine-grained emotional understanding, such as affective chat systems, moderation pipelines, and sentiment-aware conversational agents.

Model Details

Model Description

• —Developed by: Pradeep Kr. Mahato
• —Model type: Encoder-only Transformer (BERT-style)
• —Language(s): English
• —License: Apache 2.0
• —Fine-tuned from: answerdotai/ModernBERT-base
• —Task: Multi-label text classification
• —Output: Independent emotion probabilities via sigmoid activation

Model Sources

• —Base Model: https://huggingface.co/answerdotai/ModernBERT-base
• —Dataset: https://huggingface.co/datasets/google-research-datasets/go_emotions

Uses

Direct Use

The model can be used directly to:

• —Detect multiple emotions in short English texts
• —Analyze emotional tone in social media posts or chats
• —Power emotion-aware chat or moderation systems

Downstream Use

The model may be integrated into:

• —Conversational AI pipelines
• —Mental health or well-being analytics (non-diagnostic)
• —Emotion-aware recommendation systems
• —Research on affective computing

Out-of-Scope Use

The model is not intended for:

• —Medical or psychological diagnosis
• —Legal or forensic decision-making
• —High-stakes autonomous systems
• —Real-time moderation without human oversight

Bias, Risks, and Limitations

• —The training data is derived from Reddit, which may contain demographic, cultural, and topical biases
• —Emotion labels may be subjective and context-dependent
• —The model does not explicitly model emotion intensity or temporal dynamics
• —Predictions should be interpreted as probabilistic signals, not ground truth

Recommendations

• —Use human-in-the-loop validation for sensitive applications
• —Calibrate thresholds per emotion for production use
• —Avoid over-reliance on single-label interpretations

How to Get Started with the Model

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="your-username/ModernBERT-GoEmotions",
    return_all_scores=True
)

classifier("I feel anxious but also hopeful about the future.")