XiaSheng/Logic-ORiented-Retrieve
收藏Hugging Face2025-10-09 更新2025-10-25 收录
下载链接：
https://hf-mirror.com/datasets/XiaSheng/Logic-ORiented-Retrieve
下载链接
链接失效反馈官方服务：
资源简介：
Logic-ORiented Retriever Enhancement 数据集旨在用于训练和评估逻辑导向检索增强（LORE）模型。该数据集实现了三层对比学习框架，具有细粒度的样本分类：积极样本（P，标签=1），即包含足够回答查询的文本块；干扰样本（N1，标签=-1），即在大语言模型中用于查询重写看似相关但无用的文本块；负样本（N2，标签=0），即其他不相关的负文本块。
This dataset is designed for training and evaluating Logic-ORiented Retriever Enhancement (LORE) models. The dataset implements a three-tier contrastive learning framework with fine-grained sample classification: Positive (P, label=1) chunks sufficient to answer the query, Hard Negative/Distractor (N1, label=-1) chunks used by LLM in query rewriting but not helpful for answering the final query, and Negative (N2, label=0) chunks irrelevant to the query.
提供机构：
XiaSheng
