Towards Improved Classification Accuracy on Highly Imbalanced Text Dataset Using Deep Neural Language Models