Refined Commonsense Knowledge from Large-Scale Web Contents

Commonsense knowledge (CSK) about concepts and their properties is helpful for AI applications. Prior works, such as ConceptNet, have compiled large CSK collections. However, they are restricted in expressiveness to subject-predicate-object (SPO) triples with simple S strings P O. This paper presents a method called ASCENT++ automatically build large-scale base (KB) of assertions, refined both better precision recall than prior works. goes beyond SPO by capturing composite subgroups aspects, refining assertions semantic facets. The latter essential express the temporal spatial validity further qualifiers. Furthermore, combines open information extraction (OpenIE) judicious cleaning ranking typicality saliency scores. For high coverage, our taps into crawl C4 broad web contents. evaluation human judgments shows superior quality KB, an extrinsic QA-support tasks underlines benefits ASCENT++. A interface, data, code can be accessed at https://ascentpp.mpi-inf.mpg.de/.