ASER: Towards Large-scale Commonsense Knowledge Acquisition via Higher-order Selectional Preference over Eventualities

• DOI: 10.48550/arxiv.2104.02137
• Publication Date: 2021-01-01
• Authors: Hongming Zhang; Xin Liu; Haojie Pan; Haowen Ke; Jiefu Ou; Tianqing Fang; Yangqiu Song
• Keywords: Knowledge graph; Commonsense knowledge; Commonsense reasoning; Preference relation

Commonsense knowledge acquisition and reasoning have long been a core artificial intelligence problem. However, in the past, there has lack of scalable methods to collect commonsense knowledge. In this paper, we propose develop principles for collecting based on selectional preference. We generalize definition preference from one-hop linguistic syntactic relations higher-order over graphs. Unlike previous (e.g., ConceptNet), (SP) only relies statistical distribution graphs, which can be efficiently accurately acquired unlabeled corpus with modern tools. Following principle, large-scale eventuality (a term covering activity, state, event)-based graph ASER, where each is represented as dependency graph, relation between them discourse defined shallow parsing. The collected graphs reflects various kinds Moreover, motivated by observation that humans understand events abstracting observed higher level thus transfer their new events, conceptualization module significantly boost coverage ASER. total, ASER contains 648 million edges 438 eventualities. After Probase, concept-instance relational base, our concept 15 conceptualized eventualities 224 them. Detailed analysis provided demonstrate its quality. All data, APIs, tools are available at https://github.com/HKUST-KnowComp/ASER.