README.md

---
license: mit
language:
- en
library_name: spacy
tags:
- named-entity-recognition
- b2b
- ecommerce
- order-processing
- product-extraction
- spacy
datasets:
- custom
metrics:
- f1
- precision
- recall
model-index:
- name: b2b-ecommerce-ner
  results:
  - task:
      type: token-classification
      name: Named Entity Recognition
    dataset:
      type: custom
      name: B2B Ecommerce Orders
    metrics:
    - type: f1
      value: 0.82
      name: F1 Score
    - type: precision 
      value: 0.82
      name: Precision
    - type: recall
      value: 0.81
      name: Recall
---

# B2B Ecommerce NER Model

## Model Description

This is a Named Entity Recognition (NER) model specifically trained for B2B ecommerce order processing. The model extracts structured information from retailer-to-manufacturer order text, enabling automated order capture and processing.

## Supported Entities

The model identifies the following entity types:

- **PRODUCT**: Product names and descriptions (e.g., "Coca Cola", "Golden Dates", "Chocolate Cleanser")
- **QUANTITY**: Order quantities (e.g., "5", "10", "twenty")  
- **SIZE**: Product sizes and measurements (e.g., "500ML", "250G", "1.25L")
- **UNIT**: Units of measurement (e.g., "units", "bottles", "packs")

## Features

- **High Accuracy**: Achieves F1 score of 0.82 on B2B ecommerce order data
- **Product Catalog Matching**: Includes fuzzy matching against a comprehensive product catalog
- **Multi-language Support**: Handles mixed English/Hindi text common in Indian B2B commerce
- **Real-world Patterns**: Trained on actual retailer order patterns and variations

## Usage

### Basic Usage

```python
from huggingface_model.model import B2BEcommerceNER

# Load the model
model = B2BEcommerceNER.from_pretrained("path/to/model")

# Extract entities from order text
results = model.predict(["Order 5 bottles of Coca Cola 650ML"])

print(results[0])
# Output: {
#   'text': 'Order 5 bottles of Coca Cola 650ML',
#   'entities': {
#     'products': [{'text': 'Coca Cola', 'label': 'PRODUCT', 'start': 19, 'end': 28}],
#     'quantities': [{'text': '5', 'label': 'QUANTITY', 'start': 6, 'end': 7}],
#     'sizes': [{'text': '650ML', 'label': 'SIZE', 'start': 29, 'end': 34}],
#     'units': [{'text': 'bottles', 'label': 'UNIT', 'start': 8, 'end': 15}],
#     'catalog_matches': [...]
#   }
# }
```

### Pipeline Usage

```python
from huggingface_model.model import pipeline

# Create NER pipeline
ner_pipeline = pipeline("ner", model="b2b-ecommerce-ner")

# Process text
entities = ner_pipeline("I need 10 packs of biscuits")
```

### Batch Processing

```python
# Process multiple orders at once
orders = [
    "Order 5 Coke Zero 650ML",
    "Send 12 bottles of mango juice", 
    "I need 3 units of Chocolate Cleanser 500ML"
]

results = model.predict(orders)
```

## Training Data

The model was trained on a dataset of 500 B2B ecommerce orders containing:
- Real retailer-to-manufacturer communications
- Mixed English/Hindi text patterns
- Various product categories (beverages, food items, personal care)
- Different order formats and structures
- 1,002 labeled entities across 4 entity types

## Model Performance

| Metric | Score |
|--------|-------|
| F1 Score | 0.82 |
| Precision | 0.82 |
| Recall | 0.81 |

The model shows strong performance across all entity types, with particularly good results on PRODUCT and QUANTITY recognition.

## Product Catalog Integration

The model includes a fuzzy matching system that can match extracted products against a catalog of 1,855+ products, providing:

- **Brand Matching**: Match to specific brands (e.g., "Coca Cola", "Ziofit")
- **Product Variants**: Find different sizes/variants of the same product
- **Confidence Scores**: Numerical confidence for each match (0-100)
- **SKU Mapping**: Direct mapping to product SKUs for order processing

## Limitations

- Performance may vary on product names not seen during training
- Best results with English text; mixed language support is experimental
- Requires product catalog file for fuzzy matching features
- Based on spaCy framework, not transformer-based

## Technical Details

- **Framework**: spaCy 3.8+
- **Base Model**: en_core_web_sm
- **Training**: Custom NER component with 50 iterations
- **Entity Labels**: 4 custom entity types
- **Input**: Raw text strings
- **Output**: Structured entity information with optional catalog matching

## Installation

```bash
pip install spacy pandas fuzzywuzzy python-levenshtein
python -m spacy download en_core_web_sm
```

## Citation

```bibtex
@misc{b2b_ecommerce_ner_2025,
  title={B2B Ecommerce NER Model for Order Processing},
  author={Your Name},
  year={2025},
  howpublished={Hugging Face Model Hub},
  url={https://huggingface.co/your-username/b2b-ecommerce-ner}
}
```

## License

This model is released under the MIT License.