🛍️ Retail Embedding Classifier v1

This repository contains the script used to fine-tune the retail-embedding-classifier-v1, a specialized model optimized for generating embeddings for retail-specific tasks.

The fine-tuned model is available on Hugging Face.

✨ Key Features

• Product Similarity Matching
• Taxonomy Conversion
• Inventory Migration
• Grouping Similar Products
• Recommendation Systems

---

• Fine-Tuned for Retail: Trained on 217,894 samples of retail-specific product descriptions and related metadata.
• Embeddings for Retail-Specific Tasks: Generates high-quality, 768-dimensional embeddings.
• Supports Multi-Tier Retail Use Cases: From similarity searches to classification and clustering.
• Efficient Inference: Built on BAAI/bge-base-en, leveraging the CosineSimilarityLoss function for robust semantic textual similarity.

---

🚀 Quick Start

Install the required library: pip install -U sentence-transformers

Load the model and get started:

from sentence_transformers import SentenceTransformer
# Load the model
model = SentenceTransformer("mavihsrr/retail-embedding-classifier-v1")
# Example inputs
sentences = [
    "Organic Almond Butter - Creamy and Unsalted. High in Protein!",
    "Peanut Butter - All Natural, Unsalted, Smooth and Creamy.",
]
# Generate embeddings
embeddings = model.encode(sentences)
# Print shape of embeddings
print(embeddings.shape)  # Output: (2, 768)

📊 Training Details

• Base Model: BAAI/bge-base-en
• Loss Function: CosineSimilarityLoss
• Dataset: 217,894 training samples, focusing on retail-specific texts.
• Evaluation: 21,759 evaluation samples, achieving consistent high similarity scores.

💡 Use Cases

1. Product Similarity Matching Find products similar to a given item based on their embeddings:

from sklearn.metrics.pairwise import cosine_similarity
similarity_matrix = cosine_similarity(embeddings)
print(similarity_matrix)

2. Taxonomy Conversion Easily map products from one taxonomy to another using vector similarity.

3. Recommendations Build recommendation systems by clustering or ranking products based on embedding similarity.

🛠️ Advanced Configuration : Prepare Your Dataset

• Dataset Format: A .csv file with the following columns:
• sentence1: First text input.
• sentence2: Second text input (optional for single-sentence embeddings).
• label: Similarity score (range: 0 to 1). Check out the official sentence-transformer documentation for further reference.

🤝 Contributing

Contributions are welcome! If you find issues or want to improve the model, feel free to create an issue or submit a pull request.

📄 License

This project is licensed under the MIT License.

✨ Connect with Us For queries or suggestions, reach out via GitHub Issues or contact: Email: shivam.m@ionio.io