Web information extraction by competing classification

Abstract

A competing classification method is presented to extract Web information. The method uses similarity between information fragments and samples as competing ability. It classifies fragments and filters out noise information through competition of fragments for template slots. It needs far less tagged samples than those using rules to extract information. Experiments show that the method keeps high precision of information extraction without any feature clues provided by users. Therefore it is adaptive. The competing classification method is also robust in dealing with data sources having missing items and items of various orders. Show more