OKELE Dataset

该数据集是一个最近开放的知识图谱补全数据集，其中包含了从网页上抓取的含有噪声的事实以及人工标注的真实事实。该数据集重新使用了12,000个实体作为种子，收集了它们在测试集中未出现的一跳和两跳事实，以构建一个Freebase的子图。数据集涵盖了10个流行实体类别，共包含191,759条事实，每个类别分别有1,000个实体用于训练，100个实体用于验证，以及100个实体用于测试。其任务是知识图谱补全。

This dataset is a recently open-sourced knowledge graph completion dataset. It contains noise-contaminated facts scraped from web pages and manually annotated ground-truth facts. The dataset reuses 12,000 entities as seeds, and collects their one-hop and two-hop facts that are absent from the test set to construct a subgraph of Freebase. It covers 10 popular entity categories, with a total of 191,759 facts. For each category, 1,000 entities are allocated for training, 100 for validation, and 100 for testing. Its targeted task is knowledge graph completion.