Improving the classification accuracy of extremist texts on social networks by detecting and correcting grammatical errors