README.md

---
language: 
  - en
  - de
tags:
  - text-classification
  - ticket classification
  - multilingual
  - email-intent-detection
  - customer-support
  - xlm-roberta
license: apache-2.0
datasets: 
  - private
model-index:
  - name: xlm-roberta-ticket-classifier
    results:
      - task:
          type: text-classification
          name: Email Ticket Classification
        dataset:
          name: german-english-email-ticket-classification
          type: private
        metrics:
          - name: Accuracy (Type)
            type: accuracy
            value: 0.8573
          - name: Accuracy (Queue)
            type: accuracy
            value: 0.5189
          - name: F1 Score (Type)
            type: f1
            value: 0.8573
          - name: F1 Score (Queue)
            type: f1
            value: 0.5209

---

# XLM-RoBERTa Ticket Classifier

A multilingual email/ticket classifier fine-tuned from `xlm-roberta-base` to categorize customer support tickets in English and German. It predicts both routing category and issue type, helping automate ticket triage, intent detection, and prioritization in multilingual helpdesk environments.


## Model Details

- **Base model**: `xlm-roberta-base`
- **Languages**: English 🇬🇧 & German 🇩🇪
- **Task**: Multi-class text classification
- **Training data**: [german-english-email-ticket-classification](https://huggingface.co/datasets/ale-dp/german-english-email-ticket-classification)
- **Tokenizer**: SentencePiece BPE tokenizer
- **Framework**: 🤗 Transformers

## Classification Schema

This model performs **multi-head classification**, predicting both:

### 🎯 Queue (Routing Category)
- Billing and Payments  
- Customer Service  
- General Inquiry  
- Human Resources  
- IT Support  
- Product Support  
- Returns and Exchanges  
- Sales and Pre-Sales  
- Service Outages and Maintenance  
- Technical Support  

### 🛠️ Type (Issue Nature)
- Incident 
- Request  
- Problem  
- Change  

## 📈 Model Performance Summary

| Metric                  | Value   |
|-------------------------|---------|
| **Accuracy (Type)**      | 85.73%  |
| **Accuracy (Queue)**     | 51.89%  |
| **F1 Score (Type)**      | 85.73%  |
| **F1 Score (Queue)**     | 52.09%  |

This model demonstrates strong performance on **type classification**, while **queue prediction** reflects the inherent complexity of routing logic across overlapping categories.

🔍 _More detailed metrics, visualizations, and training curves available on the [W&B dashboard](https://wandb.ai/alikhalaji-/bilingual_ticket_classifier)_

## Intended Uses

- Classify incoming tickets into predefined categories
- Automate support ticket routing
- Detect customer intent in multilingual environments
- Integrate with helpdesk platforms like Zendesk or Freshdesk

## 🚀 Usage

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

model_id = "ale-dp/xlm-roberta-ticket-classifier"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

text = "Hallo, Die Data-Analytics-Plattform funktioniert nicht richtig und es werden unkorrekte Investment-Analyse-Fehlermeldungen generiert. Dies könnte auf einen Software-Fehler hindeuten."

result = classifier(text)
print(result)
```
### Created by:  
***[ᴀʟɪ ᴋʜᴀʟᴀᴊɪ](https://github.com/alikhalajii)***

## Citation

If you use this model, please cite:

```bibtex
@misc{xlm-roberta-ticket-classifier,
  author = {Ali Khalaji},
  title = {XLM-RoBERTa Ticket Classifier},
  year = {2025},
  url = {https://huggingface.co/ale-dp/xlm-roberta-ticket-classifier}
}