Evaluating NewsQA Dataset With ALBERT

Machine question answering tasks are widely in use these days. Many of the RNN (Recurrent Neural Networks) models including improved models like LSTMs (Long short-term memory), GRUs (Gated recurrent units) are being investigated to capture long term dependencies for this task. Overcoming the traditional deep learning architectures, latest transformer models outperformed the previous ones. This paper uses transformer model named ALBERT that is based on the architecture of BERT. A pre-trained ALBERT base and large models are used from the hugging face library that are fine-tuned on NewsQA dataset that is machine reading comprehension dataset. This dataset is converted into SQuAD (Stanford Question Answering Dataset) format which is later trained on our model. The model achieves state of the art results on the dataset.