Improving Vietnamese Question-Answering system with Data Augmentation and Optimization.

Extractive question answering (EQA) is one of the most important tasks in natural language processing (NLP) which has both commercial and research value. Recently, methods using neural networks, especially transformer architecture achieved state-of-the-art results in this field. Because of the rise of large language models, datasets need to be more complex to evaluate models strictly. Despite being meticulously fine-tuned on the most cutting-edge and comprehensive datasets available, these models still exhibit a surprising and concerning tendency to struggle with seemingly uncomplicated scenarios. In this paper, we propose a simple and efficient method for Extractive question answering: (1) to augment data to improve question answering task using models and original datasets; (2) we also use deque data structure to enhance post-processing process to guarantee finding the best answer and to decrease complexity to O(n) in there, n is length of context in question answering problem.