This document introduces challenges in classifying customer feedback at Cookpad, including having a large number of categories, multi-label feedback, and imbalanced data. It discusses an initial heuristic approach that was infeasible, and a second trial using simple machine learning with multiple binary SVM classifiers to handle the multi-label problem. This approach rebalanced the data and achieved high precision and recall in evaluations, reducing the time spent on feedback tagging by customer support teams by half. Future work to improve the system is also discussed.