Extracting Data Records from Unstructured Biomedical Full Text

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 837–846, Prague, June 2007. c⃝2007 Association for Computational Linguistics Extracting Data Records from Unstructured Biomedical Full Text Donghui Feng Gully Burns Eduard Hovy Information Sciences Institute University of Southern California Marina del Rey, CA, 90292 {donghui, burns, hovy}@isi.edu Abstract In this paper, we address the problem of extracting data records and their attributes from unstructured biomedical full text. There has been little effort reported on this in the research community. We argue that semantics is important for record extraction or finer-grained language processing tasks. We derive a data record template including semantic language models from unstruc- tured text and represent them with a dis- course level Conditional Random Fields (CRF) model. We evaluate the approach from the perspective of Information Extrac- tion and achieve significant improvements on system performance compared with other baseline systems. 1 Introduction The discovery and extraction of specific types of information, and its (re)structuring and storage into databases, are critical tasks for data mining, knowledge acquisition, and information integration from large corpora or heterogeneous resources (e.g., Muslea et al., 2001; Arasu and Garcia- Molina, 2003). For example, webpages of products on Amazon may contain a list of data records such as books, watches, and electronics. Automatic extraction of individual records will facilitate the access and management of data resources. Most current approaches address this problem for structured or semi-structured text, for instance, from XML format files or lists and/or tabular data records on webpages (e.g., Liu et al., 2003; Zhu et al., 2006). The techniques applied rely strongly on the analysis of document structure derived from the webpage’s html tags (e.g., the DOM tree model). Regarding unstructured text, most Information Extraction (IE) work has focused on named entities (people, organizations, places, etc.). Such IE treats each extracted element as a separate record. Much less work has focused on the case where several related pieces of information have to be extracted to jointly comprise a single data record. In this work, it is usually assumed that there is only one record for each document (e.g., Kristjannson et al., 2004). Almost no work tries to extract multiple data records from a single document. Multiple data records can be scattered across the narrative in free text. The problem becomes much harder as there are no explicit boundaries between data records and no heavily indicative format features (like html tags) to utilize. With the exponential increase of unstructured text resources (e.g., digitalized publications, papers and/or technical reports), knowledge needs have made it a necessity to explore this problem. For example, biomedical papers contain numerous ex- periments and findings. But the large volume and rate of publication have made it infeasible to read through the articles and manually identify data re- cords and attributes. We present a study to extract data records and attributes from the biomedical research literature. This is part of an effort to develop a Knowledge Base Management System …

Content truncated to 500 words