Enhancing Question Generation with Syntactic Details and Multi-Level Attention Mechanism
Cong Zhou, Jia Hu Zhu, Qing Wang, Chaojun Meng, Changfan Pan, Jianyang Shi · 2023
Generating questions is a pervasive task in natural language generation that utilizes sequence-to-sequence models based on recurrent neural networks. However, these models face a significant challenge known as the "long-term dependence" problem, which hinders their ability to capture long-term dependencies in sequences effectively. Therefore, the generated questions may not be related to the original answers or follow the correct grammar rules, making them less fluent. This paper proposes a question-generation model that combines syntactic details with an enhanced multi-level attention mechanism. Incorporating syntactic information vectors into the model’s input and using a multi-level attention mechanism to capture sequential contextual information effectively alleviates the above problems. Experimental evaluations show that the proposed model outperforms commonly used models on the SQuAD dataset.