Deep Web Classification based on Domain Feature Text

Deep web provides tremendous structured data with high quality. In order to retrieve deep web data, one important task is to classify the domains of deep web automatically. In this paper, an approach based on domain feature text (DFT) is presented to classify the deep web. In the phase of DFT selection, a semantic abstract method based on ontology knowledge and a quantitative criteria for DFT selection based on domain correlation is proposed, which enhances the representational ability of DFT and avoids the subjectivity and uncertainty of manual selection as well. In the process of the interface vector construction, an improved weighting method is given to evaluate the different roles of DFT. Finally, a KNN algorithm is used to classify these interface vectors. Experiments on 160 query interfaces in 4 typical domains demonstrate the feasibility and effectiveness of our proposed approach.