This study addresses the challenge of large-scale, multi-label recipe classification us- ing a real-world dataset of over 600,000 recipes collected from heterogeneous sources. The raw data exhibited significant noise, ...