Extracting initial strongly negative data set from positive data and unlabeled data is a base for constructing a PU-oriented text classifier by two stage frame method.