KACC: A Multi-task Benchmark for Knowledge Abstraction, Concretization and Completion

Knowledge graphs (KGs) contains an instance-level entity graph and an ontology-level concept graph. Recent studies reveal that jointly modeling of these two graphs could improve the understanding of each one. The completion processes on the concept graph and the entity graph can be further regarded as processes of knowledge abstraction and concretization. However, concept graphs in existing datasets are usually small and the links between concepts and entities are usually sparse, which cannot provide sufficient information for knowledge transfer between the two graphs. In this paper, we propose large-scale datasets extracted from Wikidata, which provide more size-balanced concept graphs and abundant cross-view links. Based on the datasets, we further propose a benchmark to test the ability of existing models on knowledge abstraction, concretization and completion (KACC). Our dataset is available at https://github.com/thunlp/kacc.