Text this: Combination of active learning and self-training for cross-lingual sentiment classification with density analysis of unlabelled samples