Text Data Mining Using Information Extraction

University Of Texas At Austin, Austin TX

Investigators

Abstract

The goal of this research project is to develop new algorithms and systems for effectively discovering knowledge in unstructured textual data. The approach first uses trained information extraction systems to obtain structured data from unstructured natural-language documents or web pages, and then applies rule induction methods to discover interesting patterns in this extracted data. Since data automatically extracted from text is noisy, heterogeneous, and non-standardized, the project studies two approaches to effectively mining extracted data. First, methods are developed for inducing rules that only partially match extracted text. Second, methods are developed that automatically cluster noisy variations of strings into standardized data items prior to mining. Algorithms are also developed for using discovered knowledge to further improve the accuracy of information extraction. Developed methods are being evaluated on large text corpora in business, medicine, science, and technology. The research will contribute to the development of technology capable of automatically discovering significant scientific, commercial, and industrial knowledge from the ever-growing supply of textual electronic information.