KACC

KACC 基准由三个可应用于知识图谱的子任务组成：知识抽象、知识具体化和知识补全。 知识抽象子任务包含两视图 KG 上的概念推理、模式预测和概念图补全任务。 知识具体化子任务需要模型基于两个子图进行实体图补全。具体化能力可以通过长尾实体链接预测的结果来进一步检验。 知识完成子任务由每个子图的典型单视图知识图完成任务组成。 KACC 在实体图中包含 999,902 个实体，有 691 种关系。概念图包含 21,293 个概念和 198 种元关系。两者之间有 2,367,971 个交叉链接。The KACC benchmark consists of three sub-tasks applicable to knowledge graphs: knowledge abstraction, knowledge concretization, and knowledge completion. The knowledge abstraction sub-task covers concept reasoning, schema prediction, and concept graph completion tasks on two-view KG. The knowledge concretization sub-task requires models to perform entity graph completion based on two subgraphs, and its concretization capability can be further evaluated via the results of long-tail entity link prediction. The knowledge completion sub-task consists of typical single-view knowledge graph completion tasks for each subgraph. The KACC benchmark contains 999,902 entities and 691 distinct relation types in the entity graph. The concept graph includes 21,293 concepts and 198 meta-relation types, with 2,367,971 cross-links between the two graphs.

搜集汇总

背景与挑战

背景概述

KACC是一个用于知识图谱的多任务基准数据集，涵盖知识抽象、具体化和补全三个子任务。该数据集包含实体图和概念图，分别具有999,902个实体与691种关系、21,293个概念与198种元关系，以及两者间的2,367,971个交叉链接，由北京邮电大学、清华大学和腾讯于2021年发布。