Open-Domain Fine-Grained Class Extraction from Web Search Queries

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 403–414, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Open-Domain Fine-Grained Class Extraction from Web Search Queries Marius Pas¸ca Google Inc. 1600 Amphitheatre Parkway Mountain View, California 94043 mars@google.com Abstract This paper introduces a method for extract- ing fine-grained class labels (“countries with double taxation agreements with india”) from Web search queries. The class labels are more numerous and more diverse than those pro- duced by current extraction methods. Also extracted are representative sets of instances (singapore, united kingdom) for the class la- bels. 1 Introduction Motivation: As more semantic constraints are added, concepts like companies become more spe- cific, e.g., companies that are in the software busi- ness, and have been started in a garage. The sets of instances associated with the classes become smaller; the class labels used to concisely describe the meaning of more specific concepts tend to be- come longer. In fact, fine-grained class labels such as “software companies started in a garage” are of- ten complex noun phrases, since they must somehow summarize multiple semantic constraints. Although Web users are interested in both coarse (e.g., “com- panies”) and fine-grained (e.g., “software compa- nies started in a garage”) class labels, virtually all class labels acquired from text by previous extrac- tion methods (Etzioni et al., 2005; Van Durme and Pas¸ca, 2008; Kozareva and Hovy, 2010; Snow et al., 2006) exhibit little syntactic diversity. Indeed, instances and class labels that are relatively com- plex nouns are known to be difficult to detect and pick out precisely from surrounding text (Downey et al., 2007). This and other challenges associated with large-scale extraction from Web text (Etzioni et al., 2011) cause the extracted class labels to usu- ally follow a rigid modifiers-plus-nouns format. The format covers nouns (“companies”) possibly pre- ceded by one or many modifiers (“software com- panies”, “computer security software companies”). Examples of actual extractions include “european cities” (Etzioni et al., 2005), “strong acids” (Pan- tel and Pennacchiotti, 2006), “prestigious private schools” (Van Durme and Pas¸ca, 2008), “aquatic birds” (Kozareva and Hovy, 2010). As an alternative to extracting class labels from text, some methods simply import them from human-curated resources, for example from the set of categories encoded in Wikipedia (Remy, 2002). As a result, class labels potentially exhibit higher syntactic diversity. The modifiers-plus-nouns for- mat (“computer security software companies”) is usually still the norm. But other formats are possi- ble: “software companies based in london”, “soft- ware companies of the united kingdom”. Vocab- ulary coverage gaps remain a problem, with many relevant class labels (“software companies of texas” “software companies started in a garage”, “soft- ware companies that give sap training”) still miss- ing. There is a need for methods that more ag- gressively identify fine-grained class labels, beyond those extracted by previous methods or encoded in existing, manually-created resources. Such class la- bels increase coverage, for example in scenarios that enrich Web search results with instances available for the …

Content truncated to 500 words