KPQDG: a key sentence prompt-tuning for automatic question-distractor pairs generation
摘要
Automatic question generation(QG) and distractor generation(DG) for reading comprehension are widely studied in the field of NLP, with application scenarios both for computer models and human education. However, most previous work treats these two tasks as separate endeavors, resulting in narrower single-objective usage scenarios. Additionally, they often rely on traditional encoder-decoder frameworks, which pose challenges for maintaining long sequence quality. To achieve the coherent generation of question-distractor pairs and enhance the quality of results to make them suitable for human examinations, we propose a novel method that utilizes key sentence information to assist in prompt-tuning the pre-trained model. Our method treats QG and DG as dual tasks and performs joint-learning to ensure semantic alignment in the generated results. Specifically, we design a graph attention encoder and a dual gated attention layer to enhance the encoding and reasoning abilities based on the source passage. Experiments show that the proposed methods align well with the characteristics of our target. Experimental results demonstrate that our model is superior in both automatic and human evaluations, with QG and DG achieving