Streamlining Assessment with Dual T5-Models For QA-Generation

This study presents a new technique for solving question-generation and answer-generation tasks. Two T5 models are used to generate question-answer pairs from a given paragraph. The first T5 model generates questions, and the second T5 model generates answers. The objective is to develop a natural technique capable of generating multiple questions and answers. A dataset is gathered with three questions and three answers for each collected paragraph. Paragraphs are collected in three domains: “Renewable Energy”,”Environmental Science”, and “Pollution”. Using the two T5 models together aims to facilitate both questioning and answering while ensuring meaningful connections between the passages, questions, and answers. Users can also answer the questions generated by the first model and compare the similarity between their answers and the model's output. An average cosine similarity score of 0.73 was obtained for the first model, while the second model achieved a cosine similarity score of 0.74.