Abstract—Generating questions from answers and articles is an interesting and difficult task. Recent works have mostly focused on quality of generating a single question from a given article. However, question generation is not only a one-to-one problem that maps an article to one single question, because everyone may focus on different parts of the article and ask different questions accordingly. This makes the diversity of generated questions equally important with the quality. In this paper, we discuss the quality and diversity of question generation, and propose a controllable question generation model to improve these two aspects. Specifically, we associate the article with the dependencies parse tree and merge it with the article by constructing different triples. Secondly, because triples are different, we can generate different questions based on different triples and articles. By selecting different triples, we can control the content of generated questions and improve both the quality and the diversity. Experiment results on the SQuAD dataset show that our proposed method can significantly improve the diversity of generated questions, especially from the perspective of using different question types. Compared with the existing methods, our model achieves a better trade-off between the quality and diversity of generated questions, and we can generate diverse questions in a more controlled way.
Index Terms—Automatic question generation, semantic graphs, pretraining language models, BART.
Zhifei Xu is with the Cornell College, China (email: ZXu22@cornellcollege.edu).
Cite: Zhifei Xu, "Controllable Question Generation with Semantic Graphs," International Journal of Machine Learning and Computing vol. 12, no. 6, pp. 351-356, 2022.
Copyright @ 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).