Ceres

Abstract

The web contains countless semi-structured websites, which can be a rich source of information for populating knowledge bases. Existing methods for extracting relations from the DOM trees of semi-structured webpages can achieve high precision and recall only when manual annotations for each website are available. Although there have been efforts to learn extractors from automatically generated labels, these methods are not sufficiently robust to succeed in settings with complex schemas and information-rich websites. In this paper we present a new method for automatic extraction from semi-structured websites based on distant supervision. We automatically generate training labels by aligning an existing knowledge base with a website and leveraging the unique structural characteristics of semi-structured websites. We then train a classifier based on the potentially noisy and incomplete labels to predict new relation instances. Our method can compete with annotation-based techniques in the literature in terms of extraction quality. A large-scale experiment on over 400,000 pages from dozens of multi-lingual long-tail websites harvested 1.25 million facts at a precision of 90%.

ZeroShotCeres: Zero-Shot Relation Extraction from Semi-Structured Webpages

“…Since this approach requires training data for each template targeted for extraction, recent work has focused on reducing the manual work needed per site. Fonduer (Wu et al, 2018) provides an interface for easily creating training data, Vertex (Gulhane et al, 2011) uses semi-supervision to minimize the number of labels needed, LODIE (Gentile et al, 2015) and Ceres (Lockard et al, 2018) automatically generate training data based on distant supervision, and DIADEM (Furche et al, 2014) identifies matching rules for specific entity types.…”

“…For this work, we assume the page topic entity has already been identified, (such as by the method proposed by Lockard et al (2018) or by using the HTML title tag) and thus limit ourselves to identifying the objects and corresponding relations. We consider the following two settings: Relation Extraction (ClosedIE): Let R define a closed set of relation types, including a special type indicating "No Relation".…”

“…To circumvent manual data annotation, previous work used a distant supervision process requiring a knowledge base aligned to the website targeted for extraction (Gentile et al, 2015;Lockard et al, 2018), including for OpenIE extraction (Banko et al, 2007;Bronzi et al, 2013;Lockard et al, 2019). These methods, however, can only learn a website-specific model based on seed knowledge for the site, but cannot be generalized to the majority of websites with knowledge from new verticals, by long-tail specialists, and in different languages.…”

FreeDOM: A Transferable Neural Architecture for Structured Information Extraction on Web Documents

“…However, in practice, active-learning methods not only constantly require human effort in building specialized annotation tools [24] but also need humans to label samples for each new site of interest. A recent work by Lockard et al [25] attempted to use additional knowledge bases as distant supervision to automatically label some samples in the target websites and then learn a machine learning model on such noisy-labeled data. A large and comprehensive knowledge base is not always accessible and available for every domain.…”