In the Rakuten data challenge on taxonomy Classification for eCommerce - scale Product Catalogs, we propose an approach based on deep convolutional neural networks to predict product taxonomies using their descriptions. The classification performance of the proposed system is further improved with oversampling, threshold moving and error correct output coding. The best classification accuracy is obtained through ensembling multiple networks trained differently with multiple inputs comprising of various extracted features.
- Convolutional neural networks
- Error correct output coding
- Imbalanced classes
- Multi-class classification
- Word embedding