Open Information Extraction Using Wikipedia

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 118–127, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Open Information Extraction using Wikipedia Fei Wu University of Washington Seattle, WA, USA wufei@cs.washington.edu Daniel S. Weld University of Washington Seattle, WA, USA weld@cs.washington.edu Abstract Information-extraction (IE) systems seek to distill semantic relations from natural- language text, but most systems use super- vised learning of relation-specific examples and are thus limited by the availability of training data. Open IE systems such as TextRunner, on the other hand, aim to handle the unbounded number of relations found on the Web. But how well can these open systems perform? This paper presents WOE, an open IE system which improves dramatically on TextRunner’s precision and recall. The key to WOE’s per- formance is a novel form of self-supervised learning for open extractors — using heuris- tic matches between Wikipedia infobox at- tribute values and corresponding sentences to construct training data. Like TextRunner, WOE’s extractor eschews lexicalized features and handles an unbounded set of semantic relations. WOE can operate in two modes: when restricted to POS tag features, it runs as quickly as TextRunner, but when set to use dependency-parse features its precision and recall rise even higher. 1 Introduction The problem of information-extraction (IE), gen- erating relational data from natural-language text, has received increasing attention in recent years. A large, high-quality repository of extracted tu- ples can potentially benefit a wide range of NLP tasks such as question answering, ontology learn- ing, and summarization. The vast majority of IE work uses supervised learning of relation- specific examples. For example, the WebKB project (Craven et al., 1998) used labeled exam- ples of the courses-taught-by relation to in- duce rules for identifying additional instances of the relation. While these methods can achieve high precision and recall, they are limited by the availability of training data and are unlikely to scale to the thousands of relations found in text on the Web. An alternative paradigm, Open IE, pioneered by the TextRunner system (Banko et al., 2007) and the “preemptive IE” in (Shinyama and Sekine, 2006), aims to handle an unbounded number of relations and run quickly enough to process Web- scale corpora. Domain independence is achieved by extracting the relation name as well as its two arguments. Most open IE systems use self- supervised learning, in which automatic heuristics generate labeled data for training the extractor. For example, TextRunner uses a small set of hand- written rules to heuristically label training exam- ples from sentences in the Penn Treebank. This paper presents WOE (Wikipedia-based Open Extractor), the first system that au- tonomously transfers knowledge from random ed- itors’ effort of collaboratively editing Wikipedia to train an open information extractor. Specifically, WOE generates relation-specific training examples by matching Infobox1 attribute values to corre- sponding sentences (as done in Kylin (Wu and Weld, 2007) and Luchs (Hoffmann et al., 2010)), but WOE abstracts these examples to relation- independent training data to learn an unlexical- …

Content truncated to 500 words