lmnop-routing-gt
收藏资源简介:
LMNOP路由基准真值数据集是一个多语言新闻事件标注数据集,专为商品领域分类任务设计。它包含421个样本,覆盖石油、液化天然气、谷物和航运四个商品领域。标注信息包括121个人工标注的基准真值和300个由Claude Sonnet 4.6模型生成的“预言机”标注。此外,数据集提供了GPT-4o模型的路由预测结果及其置信度(分为HIGH和MEDIUM级别)。每个样本包含字段:唯一标识符、新闻标题、事件批次日期、来源领域、基准真值领域、基准真值推理过程、基准真值置信度、路由预测领域和路由预测置信度。数据集仅包含一个训练集划分,适用于自然语言处理中的多标签分类、领域路由和模型预测评估任务。
The LMNOP Routing Ground-Truth Dataset is a multilingual news event annotation dataset designed for commodity domain classification tasks. It contains 421 samples specifically categorized into four commodity domains: oil, liquefied natural gas (LNG), grain, and shipping. The annotation information of this dataset originates from two sources: 121 samples are manually annotated ground-truth, while the remaining 300 samples are "oracle" annotations generated by the Claude Sonnet 4.6 model. Additionally, the dataset provides the router predictions and their confidence levels (divided into two tiers: HIGH and MEDIUM) for these samples generated by the GPT-4o model. Each data sample includes the following fields: unique identifier (id), news title (title), event batch date (date), source domains (string sequence), ground-truth domains (gt_domains, string sequence), ground-truth reasoning (gt_reasoning), ground-truth confidence (gt_confidence), router prediction domains (router_domains, string sequence), and router prediction confidence (router_confidence). The dataset only has one training split (train), with a total of 421 examples. This dataset is suitable for tasks including multi-label classification, domain routing, and model prediction evaluation in natural language processing.
数据集概述
- 数据集名称:LMNOP Routing Ground Truth
- 语言:英语(en)
- 许可协议:MIT
- 标签:commodity, routing, nlp, multi-label
- 简介:包含421条多语言新闻事件数据,用于商品领域分类(石油、液化天然气、谷物、航运)。
数据特征
| 特征名 | 数据类型 | 描述 |
|---|---|---|
| id | string | 事件唯一标识 |
| title | string | 事件标题 |
| date | string | 事件日期 |
| source_domains | 字符串序列 | 源领域 |
| gt_domains | 字符串序列 | 真实标签领域(121条手工标注 + 300条Claude Sonnet 4.6 oracle标注) |
| gt_reasoning | string | 真实标签推理过程 |
| gt_confidence | string | 真实标签置信度(HIGH / MEDIUM) |
| router_domains | 字符串序列 | GPT-4o路由器预测的领域 |
| router_confidence | string | 路由器预测置信度 |
数据集划分
| 划分 | 样本数 |
|---|---|
| train | 421 |
标注说明
- 真实标签(ground-truth labels):共421条,其中121条为手工标注,300条为Claude Sonnet 4.6 oracle标注。
- 路由器预测:由GPT-4o生成。
- 置信度等级:HIGH(高)或 MEDIUM(中)。
- 事件批次日期:Sun May 24 13:52:37 WEDT 2026。





