Reproduce “Guiding the Growth: Difficulty-Controllable Question Generation through Step-by-Step Rewriting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yi Cheng, Siyao Li, Bang Liu, Ruihui Zhao, Sujian Li, Chenghua Lin, Yefeng Zheng
A framework from Tencent Jarvis Lab, CMU, and Université de Montréal redefines question difficulty by the number of inference steps, utilizing a step-by-step rewriting approach to generate questions with controlled reasoning complexity. This method demonstrates the generation of questions with specified inference hops and boosts multi-hop QA system performance when used for data augmentation.
Key insights
- Defining question difficulty by inference steps provides a more interpretable and controllable metric than empirical or superficial features.
- A step-by-step rewriting process, guided by an explicit reasoning chain extracted from a context graph, can effectively control the logical complexity and difficulty of generated questions.
- Existing multi-hop QA datasets can be automatically repurposed to create structured training data for difficulty-controllable question generation, addressing data scarcity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.