A method on selecting reliable samples based on fuzziness in positive and unlabeled learning.