Extracting community structure features for hypertext classification