Cont information is the relation between an entity and a set of entities that can be expressed individually, as X is an employee of Z, Y is an employee of Z or as a cardinality Z has N employees. Identifying such information in text and knowledge bases (KBs) and linking the two variants can
- identify incomplete entity enumerations
- ground counts in representative entities
- identify KB inconsistencies
- enhance downstream tasks like search and question answering
Most research focus on numerical facts without joining the dots between numerical facts specifying cardinal information and their corresponding entity-to-entity grounding facts. Popular general-purpose KBs have limited ability to recognize such predicates.
We first tackle the problem of class cardinality comparison by identifying three related problems of varying informativeness and approachability.
- Cardinality estimation
- Proportionality etimation
- Dominance estimation
We show how using an aggregate of different cardinality signals of different types and from different sources, we are able to better identify the bigger of two classes.
6. Shrestha Ghosh, Simon Razniewski, Gerhard Weikum. Class Cardinality Comparison as a Fermi Problem. (WWW 2023) [pdf] [poster]
2. Shrestha Ghosh, Simon Razniewski, Gerhard Weikum. Uncovering Hidden Semantics of Set Information in Knowledge Bases. Journal of Web Semantics (JWS 2020) [pdf]