Visual HTML Document Modeling for Information Extraction

HTML, Information Extraction, Document Modeling, Logical Document Structure, Visual Information

Current methods for the information extraction from HTML documents are mostly based on wrappers that read the HTML code and identify the data to be extracted by some properties of the surrounding HTML tags and the text. The bottleneck of this approach is too tight binding of the wrapper to the HTML code. The nature of HTML allows to achieve the desired document design by various ways that can be arbitrarily combined, which makes the wrappers limited to a narrow set of documents and a short time period. By contrast, there exist some generally accepted rules for the visual data presentation in the documents. Our approach is based on using the visual information for identifying the data in the documents. We define formal models of the visual information and we propose the method for information extraction based on the unordered tree matching algorithms.