Scaling up Open Tagging from Tens to Thousands: Comprehension Empowered Attribute Value Extraction from Product Title

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5214–5223 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 5214 Scaling Up Open Tagging from Tens to Thousands: Comprehension Empowered Attribute Value Extraction from Product Title Huimin Xu1,2, Wenting Wang2, Xin Mao1,2, Xinyu Jiang1, Man Lan1∗ 1 School of Computer Science and Software Engineering, East China Normal University 2 Alibaba Group {hmxu,xinmao,xyjiang}@stu.ecnu.edu.cn, mlan@cs.ecnu.edu.cn {xinjiu.xhm,nantiao.wwt,sunian.mx}@alibaba-inc.com Abstract Supplementing product information by ex- tracting attribute values from title is a crucial task in e-Commerce domain. Previous studies treat each attribute only as an entity type and build one set of NER tags (e.g., BIO) for each of them, leading to a scalability issue which unfits to the large sized attribute system in real world e-Commerce. In this work, we propose a novel approach to support value extraction scaling up to thousands of attributes without losing performance: (1) We propose to regard attribute as a query and adopt only one glob- al set of BIO tags for any attributes to reduce the burden of attribute tag or model explosion; (2) We explicitly model the semantic repre- sentations for attribute and title, and develop an attention mechanism to capture the interac- tive semantic relations in-between to enforce our framework to be attribute comprehensive. We conduct extensive experiments in real-life datasets. The results show that our model not only outperforms existing state-of-the-art N- ER tagging models, but also is robust and gen- erates promising results for up to 8, 906 at- tributes. 1 Introduction Product attributes are vital to e-Commerce as plat- forms need attribute details to make recommen- dations and customers need attribute information to compare products and make purchase decision- s. However, attribute information is often noisy and incomplete because of the inevitable hurdles posed to retailers by the extremely huge and com- plex e-Commerce attribute system. On the oth- er hand, product titles which are carefully de- signed by retailers are packed tightly with detail- s to highlight all important aspects of product- s. Figure 1 shows the product page of a ‘dress’ from AliExpress1 which is an emerging and fast- 1https://www.aliexpress.com/ Figure 1: Snapshot of a product page. growth global e-Commerce platform. The prod- uct title “2019 Summer Women Button Decorat- ed Print Dress Off-shoulder Party Beach Sundress Boho Spaghetti Long Dresses Plus Size FICUS- RONG” contains attribute values: (1) already list- ed in Item Specifics, such as ‘Women’ for Gen- der, ‘Summer’ for Season, etc; (2) missing in Item Specifics, such as ‘2019’ for Year, ‘Plus Size’ for Size, etc. In this paper, we are interested in sup- plementing attribute information from product ti- tles, especially for the real world e-Commerce at- tribute system with thousands of attributes built-in and new attributes and values popping out every- day. Previous work (Ghani et al., 2006; Ling and Weld, 2012; Sheth et al., 2017) on attribute value extraction suffered from Closed World Assump- tion which heavily depends on certain pre-defined attribute value vocabularies. These methods …

Content truncated to 500 words