TREC-DL (2019-2022)

该数据集名为TREC-DL，包含了文档和查询，以及针对查询-文档对的由人工标注的相关性判断。查询被分为训练集、验证集和测试集，并生成了50个随机分割以进行交叉验证。该设置涉及一个前1000名的重新排序任务。数据规模涵盖了多个查询和文档，并带有相关性判断。任务的目的是进行查询-文档相关性预测。

The dataset named TREC-DL comprises documents, queries, and human-annotated relevance judgments for query-document pairs. Queries are split into training, validation, and test sets, with 50 random splits created for cross-validation purposes. This setup targets a top-1000 re-ranking task. The dataset encompasses multiple queries and documents paired with their respective relevance annotations, and the core task of this dataset is query-document relevance prediction.

搜集汇总

背景与挑战

背景概述

该数据集是Google DeepMind发布的复合检索系统代码库，包含基线方法和新方法的实现，与SIGIR 2025发表的论文相关，采用Apache 2.0和CC-BY许可证。