Machine Learning for Coding Retail Product Names to Consumer-Price Categories: A Rule-plus-Bag-of-Words Pipeline with Reliability-Weighted Human-in-the-Loop Labeling
The study created a hybrid system to map messy retail product names to price categories using rules, machine learning, and human labeling. The best result came from a simple character n-gram logistic regression, showing that complex models like CNNs and LSTMs performed worse on small datasets. The pipeline ensures auditable classification through rule-based steps and quality control.