Intent Detection for Finance - fin-customer-support-intent-distilbert

Overview

This model is a DistilBERT-based intent classifier fine-tuned on the PolyAI/banking77 dataset. It is designed to identify customer support intents from short text queries, enabling automated routing and response systems in financial and service-oriented applications. The model is optimized for both performance and efficiency, making it suitable for real-time use cases such as chatbots, virtual assistants, and support automation pipelines.

Model Details

• Architecture: DistilBERT
• Task: Multi-class text classification (77 intents)
• Dataset: PolyAI/banking77
• Language: English

Performance

The model demonstrates strong generalization on unseen data:

• Accuracy: ~0.92
• Precision: ~0.93
• Recall: ~0.92
• F1 Score: ~0.92

These results indicate reliable performance across a wide range of customer queries.

Intended Use

This model is designed for:

• Customer support automation
• Intent detection in conversational systems
• Query routing and workflow triggering
• Integration into AI agents or backend services

It can be used as a standalone classifier or as part of a larger pipeline that includes business logic and response generation.

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="mr-checker/fin-customer-support-intent-distilbert"
)

result = classifier("I lost my debit card")
print(result)

Label Mapping

The model outputs labels in the format LABEL_X.

A label_mapping.json file is provided to convert these into human-readable intent names.

Limitations

• Trained on banking-related queries; may not generalize well to unrelated domains
• Performance may drop for very long or ambiguous inputs
• Does not capture user emotion or sentiment

---

License and Acknowledgements

This model is released under the MIT License. The training dataset (PolyAI/banking77) is also licensed under MIT.

Acknowledgements:

• PolyAI for the BANKING77 dataset
• Hugging Face for the Transformers ecosystem