A Novel Methodology for Developing Automatic Harassment Classifiers for Twitter