Weak Positive Sampling and Soft Smooth Labeling for Distractor Generation Data Augmentation

Distractor generation is one of the most important and challenging tasks in the automatic generation of multiple choice questions. Previous studies usually use a few ground truth distractors as training samples, which ignores more potential usable distractors, where the strong generation ability of deep learning models might not be fully utilized. Therefore, we propose a data augmentation framework for distractor generation, which first applies the distractor ranking model on a distractor candidate set and then selects useful distractor candidates as additional training samples. Besides, we propose weak positive sampling and soft smooth labeling mechanism to ensure the sample quality and effectively use samples during the training stage. Experimental results on public benchmarks demonstrate the effectiveness of our proposed method.