Diploma Thesis Analysis and Comparison of Existent Information Extraction Methods

Information extraction is initially applied for identification of desired information from natural language documents and conversion of the extracted text into a self-defined presentation. With the rapidly increasing amount of available information sources and electronic documents on the World Wide Web, information extraction is extended for identification from structured and semi-structured web pages. In the past years, a lot of solutions are described and various information extraction systems are implemented. In this thesis, we present a theoretical analysis and comparison of several information extraction systems in two sub-areas: on the one hand, the systems distinguish from each other in the features used for identification. Thereby, we compare several aspects of different information extraction systems, such as pre-processing for generation of features, various constraints for characterization of target information, and different representations of extraction patterns, namely, how constraints are utilized. On the other hand, in order to reduce human efforts and improve the portability of information extraction systems, diverse machine learning techniques are applied for building information extraction systems. In this thesis, we represent various types of training data and introduce the active learning technique, the boosting algorithm and different rule learning algorithms used in information extraction systems. Most of the information extraction systems mentioned in this thesis employ rule learning techniques. Thereby, the structure, the evaluation heuristics and the pruning methods are compared. In addition, bayesian learning applied in information extraction system is presented as well.