samurai9776/thought-classifier-menu-aware

Thought Completion Classifier - Menu Aware v2

A fine-tuned DistilBERT model that classifies whether a customer's response in a conversation represents a complete or incomplete thought.

Model Details

• Base Model: DistilBERT
• Task: Binary Classification (Complete/Incomplete)
• Training Data: 15,000+ examples
• Optimization: High precision for incomplete thought detection

Usage

from transformers import pipeline

# Load the model
classifier = pipeline("text-classification", model="samurai9776/thought-classifier-menu-aware")

# Example usage
result = classifier("What sandwich would you like? [SEP] Chicken")
print(result)
# [{'label': 'COMPLETE', 'score': 0.98}]

result = classifier("Anything else? [SEP] Chicken")
print(result)  
# [{'label': 'INCOMPLETE', 'score': 0.95}]

Performance

• Overall Accuracy: ~92%
• Handles menu ambiguities (e.g., "chicken" appears in 75 menu items)
• Context-aware predictions

Training Details

• Fixed 3,000+ mislabeled examples from original dataset
• Fine-tuned from samurai9776/thought-classifier
• Optimized for restaurant/QSR ordering scenarios