Solving Price Per Unit Problem Around the World: Formulating Fact Extraction as Question Answering

Price Per Unit (PPU) is an essential information for consumers shopping on\ne-commerce websites when comparing products. Finding total quantity in a\nproduct is required for computing PPU, which is not always provided by the\nsellers. To predict total quantity, all relevant quantities given in a product\nattributes such as title, description and image need to be inferred correctly.\nWe formulate this problem as a question-answering (QA) task rather than named\nentity recognition (NER) task for fact extraction. In our QA approach, we first\npredict the unit of measure (UoM) type (e.g., volume, weight or count), that\nformulates the desired question (e.g., "What is the total volume?") and then\nuse this question to find all the relevant answers. Our model architecture\nconsists of two subnetworks for the two subtasks: a classifier to predict UoM\ntype (or the question) and an extractor to extract the relevant quantities. We\nuse a deep character-level CNN architecture for both subtasks, which enables\n(1) easy expansion to new stores with similar alphabets, (2) multi-span\nanswering due to its span-image architecture and (3) easy deployment by keeping\nmodel-inference latency low. Our QA approach outperforms rule-based methods by\n34.4% in precision and also BERT-based fact extraction approach in all stores\nglobally, with largest precision lift of 10.6% in the US store.\n