MReD

MReD是一个针对结构可控文本生成的新型数据集，由阿里巴巴达摩院创建。该数据集包含7,089个元评论及其45,000个句子，每个句子都经过人工标注，属于9个精心定义的类别之一。MReD数据集的创建旨在深入理解元评论的结构，使模型能够根据输入文本和控制信号生成符合特定结构的元评论。该数据集的应用领域主要集中在提高文本生成的灵活性和实用性，特别是在学术论文的同行评审系统中，帮助区域主席根据不同的控制信号生成不同结构的元评论。

MReD is a novel dataset for structured controllable text creation, developed by Alibaba DAMO Academy. This dataset includes 7,089 meta-reviews and 45,000 corresponding sentences, each of which has been manually annotated into one of nine meticulously defined categories. The MReD dataset is designed to achieve in-depth understanding of the structure of meta-reviews, allowing models to generate meta-reviews with specific structures based on input texts and control signals. Its application focuses on improving the flexibility and practicality of text generation, especially in academic paper peer review systems, where it assists area chairs in generating meta-reviews of different structures based on varied control signals.