@online{Ghosh_arXiv2003.03155,
TITLE = {Uncovering Hidden Semantics of Set Information in Knowledge Bases},
AUTHOR = {Ghosh, Shrestha and Razniewski, Simon and Weikum, Gerhard},
LANGUAGE = {eng},
URL = {http://arxiv.org/abs/2003.03155},
EPRINT = {2003.03155},
EPRINTTYPE = {arXiv},
YEAR = {2020},
ABSTRACT = {Knowledge Bases (KBs) contain a wealth of structured information about<br>entities and predicates. This paper focuses on set-valued predicates, i.e., the<br>relationship between an entity and a set of entities. In KBs, this information<br>is often represented in two formats: (i) via counting predicates such as<br>numberOfChildren and staffSize, that store aggregated integers, and (ii) via<br>enumerating predicates such as parentOf and worksFor, that store individual set<br>memberships. Both formats are typically complementary: unlike enumerating<br>predicates, counting predicates do not give away individuals, but are more<br>likely informative towards the true set size, thus this coexistence could<br>enable interesting applications in question answering and KB curation.<br> In this paper we aim at uncovering this hidden knowledge. We proceed in two<br>steps. (i) We identify set-valued predicates from a given KB predicates via<br>statistical and embedding-based features. (ii) We link counting predicates and<br>enumerating predicates by a combination of co-occurrence, correlation and<br>textual relatedness metrics. We analyze the prevalence of count information in<br>four prominent knowledge bases, and show that our linking method achieves up to<br>0.55 F1 score in set predicate identification versus 0.40 F1 score of a random<br>selection, and normalized discounted gains of up to 0.84 at position 1 and 0.75<br>at position 3 in relevant predicate alignments. Our predicate alignments are<br>showcased in a demonstration system available at<br>https://counqer.mpi-inf.mpg.de/spo.<br>},
}