Mitigating shortage of labeled data using clustering-based active learning with diversity exploration