Research on Question Generation Based on Free-text

Question generation refers to automatically generating reasonable and relevant questions based on given texts and answers. Past research on question generation has suffered from the following deficiency: in question generation from free texts, the ability to extract semantic information from long texts and answer information is inadequate, resulting in the inability to obtain rich information features. To address this shortcoming, an answer-dependent annotation method was proposed in this paper, which annotates text related to the answer to fully expand the information of the answer in the original sentence and improve the matching degree between questions and answers. Meanwhile, an attention mechanism combining answers and texts was designed in the encoder to enhance answer features by calculating the attention values between answers and texts. A copying mechanism was incorporated into the decoder to strengthen the model's ability to collect text information, improving the quality of questions generated by the model. The method designed in this paper improved BLEU-4 by 0.31% on the SQuAD dataset.