mteb/KnowMeBench

该数据集是一个用于文本检索和多选问答任务的英语数据集，源自KaLM-Embedding/LMEB。它包含三个主要主题（事件驱动、倒叙密集和心理深度）和五个子任务（对抗性弃权、信息提取、身心交互、记忆触发分析和时间推理）。每个子任务配置包括语料库（含ID、文本和标题）、查询（含ID和文本）和相关性评分（含查询ID、语料库ID和分数）。所有数据仅提供测试集，用于评估检索系统在不同复杂叙事和推理场景下的性能。This dataset is an English dataset for text retrieval and multiple-choice QA tasks, derived from KaLM-Embedding/LMEB. It includes three main themes (event-driven, flashback-intensive, and psychological depth) and five subtasks (adversarial abstention, information extraction, mind-body interaction, mnestic trigger analysis, and temporal reasoning). Each subtask configuration consists of a corpus (with ID, text, and title), queries (with ID and text), and relevance judgments (with query-id, corpus-id, and score). All data is provided only in test splits, designed to evaluate retrieval systems in complex narrative and reasoning scenarios.