Lather, Rinse, Receipt

We used a donut model to extract receipt features. We matched the highest fuzzy ratio between test data rows and after receipt features were concatenated as one string with over 80% accuracy.