An Automated Domain Porting Toolkit for Information Extraction

Information extraction (IE) systems assist analysts to assimilate information from electronic documents. This task will develop an automated, domain porting toolkit for customizing a generic IE system for a specific domain. Customization is required atvarious levels: (i) lexicon, (ii) glossaries of names of people, organizations, locations, (iii) relationships between key entities, e.g. an organization's headquarters, and (iv) significant events, e.g. transportation of chemicals.Due to performance gained through customization, many have developed handcrafted IE systems applicable to a single domain, e.g. insurance and medical. The approach proposed builds on a robust, domain-independent IE engine that can continue to be enhanced.Using this as a core, we propose to develop a platform for automated domain or corpus customization. This approach facilitates rapid domain porting and cost savings since linguists are not required.The domain porting toolkit requires research in unsupervised machine learning. Structure-based training leverages output from the core IE engine. A bootstrap approach using seeds is proposed. Lexicons and templates are learned to customize all levels ofIE.Deliverables include: (i) algorithms for structure-based bootstrap learning, (ii) prototype for domain porting of lexicons and rule templates for the intelligence domain, and (iii) an automated domain porting toolkit, including graphical interfaces.