QUESTION ANSWERING SYSTEM UPON UNIFIED LANGUAGE MODEL AND EVALUATING PERFORMANCE OF DATASETS

Present days require automation and optimization in simplebut urgent tasks. It is granted to use opportunities of technologies and science inorder to work efficiently and to stay productive. In this paper, I seek tounderstand opportunities and drawbacks of the publicly available datasets, suchas SQuAD, TriviaQA, Natural Questions (NQ), QuAC, NewsQA. It is vital tochoose a suitable dataset in order to create a system with better performance.Specifically, the paper proposes an automatic question creating system that usesstate-of-the-art Natural Language Processing (NLP) - Unified Language Model(UniLM). The question generating algorithm was verified using best datasets,and it has shown noteworthy results - questions generated were logical andcorrect. This study is important for teachers, teacher assistants, to save timewriting test questions and spend it for more important duties.