Deep hierarchical product classification based on pre-trained multilingual knowledge

The customer experience of online shopping is largely contingent on the accuracy of product classification. Considering the amount of products and all the possible categories, it is desirable to construct a framework to auto-assign products into correct categories at scale. Machine learning based systems often suffer from poor data quality, such as incomplete item descriptions, adversarial noise in the training data, etc., causing low precision/recall of predictions. To overcome these difficulties, we propose a deep hierarchical product classifier based on BERT pretrained knowledge. Additionally, we propose several learning strategies, e.g., bootstrap learning, negative sampling, soft label and semantic augmentation, to capture consistent knowledge hidden behind noisy data to prevent overfitting. Experiments on a large data set with different data configurations prove the effectiveness of the proposed model.