ATOMIC is a large-scale commonsense knowledge graph (CSKG) containing
everyday if-then knowledge triplets, i.e., {head event, relation, tail event}.
The one-hop annotation manner made ATOMIC a set of independent bipartite
graphs, which ignored the numerous missing links between events in different
bipartite graphs and consequently caused shortcomings in knowledge coverage and
multi-hop reasoning. To address these issues, we propose a CSKG completion
approach by training a relation prediction model based on a set of existing
triplets, and infer the missing links on ATOMIC. On this basis, we construct
Dense-ATOMIC, a densely-connected and multi-hop commonsense knowledge graph.
The experimental results on an annotated dense subgraph demonstrate the
effectiveness of our CSKG completion approach upon ATOMIC. The evaluation on a
downstream commonsense reasoning task also proves the advantage of Dense-ATOMIC
against conventional ATOMIC.

本文提出了一种基于关系预测模型的常识知识图谱完备性方法，用于填补常识知识图谱中双方之间缺失的知识链接，并构建了一种密集连接和多跳的常识知识图谱 Dense-ATOMIC，实验证明该方法提高了 ATOMIC 的效率。