EQG-RACE-Diff-Dis for Difficulty-Controllable Question-Distractor pairs Generation

The EQG-RACE-Diff-Dis dataset can be used for difficulty-controllable question generation. We employ Bloom's theory to define question difficulty levels. The core of Bloom's theory lies in its progression from lower-order to higher-order thinking skills. Each question type in Bloom's taxonomy can be categorized into three common difficulty levels: easy, medium, and hard. Here, \easy\ corresponds to \recognition\ and \recall\ in Bloom's theory, \medium\ corresponds to \understanding,\ and \hard\ corresponds to the other classifications. Based on this, we adopt the LLM-AAL method to annotate the difficulty levels of EQG-RACE.