metadata

datasets:
  - leonvanbokhorst/topic-drift

Topic Drift Detector Model

Version: v20241225_085248

This model detects topic drift in conversations using an enhanced attention-based architecture.

Model Architecture

Multi-head attention mechanism
Bidirectional LSTM for pattern detection
Dynamic weight generation
Semantic bridge detection

Performance Metrics

=== Full Training Results ===
Best Validation RMSE: 0.0107
Best Validation R²: 0.8867

=== Test Set Results ===
Loss: 0.0002
RMSE: 0.0129
R²: 0.8373

Training Curves

Usage

import torch

# Load model
model = torch.load('models/v20241225_085248/topic_drift_model.pt')

# Use model for inference
# Input shape: [batch_size, sequence_length * embedding_dim]
# Output shape: [batch_size, 1] (drift score between 0 and 1)