Knowledge-Driven Distractor Generation for Cloze-Style Multiple Choice Questions

In this paper, we propose a novel configurable framework to automatically generate distractive choices for open-domain cloze-style multiple-choice questions. The incorporates general-purpose knowledge base effectively create small distractor candidate set, and feature-rich learning-to-rank model select distractors that are both plausible reliable. Experimental results on new dataset across four domains show our yields outperforming previous methods by automatic human evaluation. can also be used as benchmark generation research in the future.