Believing Finite-State cascades in Knowledge-based Information Extraction
Benjamin Adrian; Andreas Dengel
In: Andreas Dengel; Karsten Berns; Thomas Breuel; Frank Bomarius; Thomas Roth-Berghofer (Hrsg.). KI 2008: Advances in Artificial Intelligence. German Conference on Artificial Intelligence (KI), Kaiserslautern, Germany, Pages 152-159, Lecture Notes in Computer Science (LNCS), Vol. 5243, ISBN 978-3-540-85844-7, Springer, 2008.
Common information extraction systems are built upon regular extraction patterns and finite-state transducers for identifying relevant bits of information in text. In traditional systems a successful pattern match results in populating spreadsheet-like templates formalizing users` information demand. Many IE systems do not grade extraction results along a real scale according to correctness or relevance. This leads to difficult management of failures and missing or ambiguous information. The contribution of this work is applying belief of Dempster-Shafer´s A Mathematical Theory of Evidence for grading IE results that are generated by probabilistic FSTs. This enhances performance of matching uncertain information from text with certain knowledge in knowledge bases. The use of belief increases precision especially in modern ontology-based information extraction systems.