DMIR01/DMRetriever_MTT

DMRetriever-MTT数据集是一个大规模文本三元组（MTT）数据集，用于训练DMRetriever模型。该数据集通过大规模的文本对生成、基于互同意的假阳性过滤和难度感知的硬负样本挖掘构建而成，用于提高灾害管理中的文本检索性能。数据集包含1,137,630个样本，版本为MTT-0.85，难度水平设置为0.85。DMRetriever-MTT is a Massive Text Triplets (MTT) dataset constructed for training the DMRetriever model. It is built through a large-scale pipeline of text-pair generation, mutual-agreement–based false-positive filtering, and difficulty-aware hard-negative mining for improved text retrieval in disaster management. The dataset contains 1,137,630 samples and is released in the MTT-0.85 version with a difficulty level set to 0.85.