Bootstrapped Named Entity Recognition for Product Attribute Extraction

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1557–1567, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Bootstrapped Named Entity Recognition for Product Attribute Extraction Duangmanee (Pew) Putthividhya eBay Inc. 2065 Hamilton Ave San Jose, CA 95125 dputthividhya@ebay.com Junling Hu eBay Inc. 2065 Hamilton Ave San Jose, CA 95125 juhu@ebay.com Abstract We present a named entity recognition (NER) system for extracting product attributes and values from listing titles. Information extrac- tion from short listing titles present a unique challenge, with the lack of informative con- text and grammatical structure. In this work, we combine supervised NER with bootstrap- ping to expand the seed list, and output nor- malized results. Focusing on listings from eBay’s clothing and shoes categories, our bootstrapped NER system is able to identify new brands corresponding to spelling variants and typographical errors of the known brands, as well as identifying novel brands. Among the top 300 new brands predicted, our system achieves 90.33% precision. To output normal- ized attribute values, we explore several string comparison algorithms and found n-gram sub- string matching to work well in practice. 1 Introduction Traditional named entity recognition (NER) task has expanded beyond identifying people, location, and organization to book titles, email addresses, phone numbers, and protein names (Nadeau and Sekine 2007). Recently there has been a surge of interest in extracting product attributes from online data due to the rapid growth of E-Commerce. Current work in this domain focuses on mining product reviews and descriptions from retailer websites. Such text data tend to be long and generate enough context for the target task (Brody and Elhadad 2010; Liu et al. 2005; Popescu and Etzioni 2005). In this paper, we focus on mining short product listing titles, which poses unique challenges. Short listings are typical in classified ads where each seller is given limited space (in terms of words) to describe the product. On eBay, product listing ti- tles cannot exceed 55 characters in length. Similarly, on Craigslist and newspaper ads, the length of a list- ing title is restricted. Extracting product attributes from such short titles faces the following challenges: • Loss of grammatical structure in short listings where many nouns are piled together. • Typographical errors, abbreviations, and acronyms that must be normalized to the standardized values. • Lack of contextual information to infer product attribute value. It can be argued that the use of short listings simpli- fies the problem of attribute extraction, since short listings can be easily annotated and one can apply supervised learning approach to extract product at- tributes. However, as the size of the data grows, ob- taining labeled training set on the scale of millions of listings becomes very expensive. In such a sce- nario, incorporating unlabeled examples in a semi- supervised fashion to scale up the solution becomes a necessity rather than a luxury. We formulate the product attribute extraction problem as a named entity recognition (NER) task and investigate supervised and semi-supervised ap- proaches …

Content truncated to 500 words