|
|
--- |
|
|
language: "no" |
|
|
tags: |
|
|
- token-classification |
|
|
- text-classification |
|
|
- ner |
|
|
- intent-classification |
|
|
- norwegian |
|
|
- bert |
|
|
base_model: NbAiLab/nb-bert-large |
|
|
--- |
|
|
|
|
|
# Joint Intent + NER Model for Norwegian Feed Orders |
|
|
|
|
|
A joint intent classification and named entity recognition model for Norwegian animal feed order queries, fine-tuned from [NbAiLab/nb-bert-large](https://huggingface.co/NbAiLab/nb-bert-large). |
|
|
|
|
|
## Task |
|
|
|
|
|
Given a Norwegian spoken order query, the model simultaneously: |
|
|
- **Classifies the intent** (5 classes): `create_order`, `edit_order`, `confirm`, `reject`, `help` |
|
|
- **Extracts named entities** (7 entity types, IOB2): `PRODUCT`, `QUANTITY`, `UNIT`, `DELIVERY_METHOD`, `DELIVERY_DATE`, `ADDRESS`, `TANK_SILO` |
|
|
|
|
|
## Test Set Results |
|
|
|
|
|
| Metric | Score | |
|
|
|--------|-------| |
|
|
| NER Precision | 92.71% | |
|
|
| NER Recall | 95.19% | |
|
|
| NER F1 | 93.93% | |
|
|
| Intent Accuracy | 100% | |
|
|
| Intent F1 | 100% | |
|
|
| **Combined F1** | **96.97%** | |
|
|
|
|
|
## Training |
|
|
|
|
|
- **Base model**: NbAiLab/nb-bert-large (1024 hidden, 24 layers) |
|
|
- **Training data**: 450 examples (train + val merged after HP search) |
|
|
- **Hyperparameters**: Found via Optuna search (20 trials), then retrained on train+val |
|
|
- **Loss**: `0.6 * intent_loss + 0.4 * ner_loss` |
|
|
- **Epochs**: 5 |
|
|
- **Learning rate**: 5e-05 |
|
|
- **Batch size**: 8 |
|
|
|
