hr-engagement-classifier

Employee Sentiment Classifier — DistilRoBERTa (Fine-tuned)

This model is a fine-tuned version of j-hartmann/emotion-english-distilroberta-base on a custom HR feedback dataset containing employee survey responses.

It is designed to classify text responses into the following sentiment categories:

• Disengaged
• Content
• Engaged
• At Risk of Leaving

Model Details

• Base Model: distilroberta-base (via j-hartmann's emotion model)
• Fine-tuned on: Employee survey feedback
• Framework: Hugging Face Transformers
• Training: Multi-class classification with W&B sweeps for hyperparameter tuning

Labels

Evaluation Metrics

Evaluated on a held-out test set of employee reviews:

• Accuracy: 91.75%
• Macro F1 Score: 91.69%
• Eval Loss: 0.380

These metrics indicate strong generalization on multi-class sentiment prediction in real HR text data.

💡 Intended Use

This model is intended for analyzing internal employee sentiment from free-text responses, especially for HR and PeopleOps use cases (e.g. engagement surveys, exit feedback, etc.)

👩‍💻 Author

Fine-tuned and maintained by @yaagni