Dissertation Thesis

Information Extraction from HTML Documents Based on Logical Document Structure

Radek Burget
burgetr@fit.vutbr.cz

Abstract

The World Wide Web presents the largest Internet source of information from a broad range of areas. The web documents are mostly written in the Hypertext Markup Language (HTML) that doesn't contain any means for semantic description of the content and thus the contained information cannot be processed directly. Current approaches for the information extraction from HTML are mostly based on wrappers that identify the desired data in the document according to some previously specified properties of the HTML code. The wrappers are limited to a narrow set of documents and they are very sensitive to any changes in the document formatting.

In this thesis, we propose a novel approach to information extraction that is based on modeling the visual appearance of the document. We show, that there exist some general rules for the visual presentation of the data in documents and we define formal models of the visual information contained in a document. Furthermore, we propose the way of modeling the logical structure of an HTML document based on the visual information. Finally, we propose methods for using the logical structure model for the information extraction task based on tree matching algorithms. The advantage of this approach is certain independence on the underlying HTML code and better resistance to changes in the documents.

Full Text

Short version: burget_thesis_short.pdf (last update Aug 24, 2004)
Full version: burget_thesis.pdf or burget_thesis.ps (last update Aug 24, 2004)