New EXPONA framework automates data labeling by intelligently generating and balancing multiple label functions across surface, structural, and semantic levels.

EXPONA addresses the high cost and error-prone nature of manual data annotation by automating label function generation for weak supervision

The framework systematically explores multi-level label functions spanning surface-level, structural, and semantic perspectives to improve coverage

EXPONA implements reliability-aware mechanisms to filter out noisy label functions and maintain high-quality training data

Approach balances diversity and reliability in label generation, moving beyond limitations of existing LLM-based or hand-crafted primitive synthesis methods