Vladimir Beskorovainyi

The paper proposes a robust, multi-stage pipeline combining rule-based classification and machine learning to map noisy retail product names to standardized consumption categories, finding that simple models achieve near-perfect accuracy with minimal labeled data.