Question generation based on knowledge base

Question generation refers to automatically generating reasonable and relevant questions based on given text and answers. The application prospects of question generation technology are very broad and it has become one of the research hotspots in the field of natural language processing. In knowledge-based question generation, most models only focus on the answer entity in the knowledge base and ignore other information related to the entity, while many models are not adaptable to out-of-vocabulary predicates. To address this shortcoming, this paper proposes a controlled generation framework based on a knowledge base, which controls the generator through constraints to ensure that the given answers and key entities in the subgraph appear in the generated questions. At the same time, an optimization strategy for the pre-trained generation model is proposed, which first adjusts the parameters in the decoder and then freezes the parameters in the encoder, so that the model can retain the characteristics of the hidden predicates in the pre-trained model, thereby improving the model's adaptability to out-of-vocabulary predicates and improving the quality of generated questions.