Dense-ATOMIC: Construction of Densely-connected and Multi-hop Commonsense Knowledge Graph upon ATOMIC

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.