Open Knowledge Enrichment for Long-tail Entities

Abstract

Knowledge bases (KBs) have gradually become a valuable asset for many AI applications. While many current KBs are quite large, they are widely acknowledged as incomplete, especially lacking facts of long-tail entities, e.g., less famous persons. Existing approaches enrich KBs mainly on completing missing links or filling missing values. However, they only tackle a part of the enrichment problem and lack specific considerations regarding long-tail entities. In this paper, we propose a full-fledged approach to knowledge enrichment, which predicts missing properties and infers true facts of long-tail entities from the open Web. Prior knowledge from popular entities is leveraged to improve every enrichment step. Our experiments on the synthetic and real-world datasets and comparison with related work demonstrate the feasibility and superiority of the approach.

RETA: A Schema-Aware, End-to-End Solution for Instance Completion in Knowledge Graphs

“…KG completion problems have also been studied as a relation prediction task, suggesting missing relations to a given entity [1,2,7,10,24,52]. Specifically, for a given entity, the objective is to suggest a list of relations (so-called properties by some of these works) which are relevant to the entity.…”

“…Lajus et al [24] studied the problem of determining obligatory relations for a given entity in a KG by extracting and using the class hierarchy (of entities) in the KG. Cao et al [7] designed a relation prediction technique by applying an attention-based graph neural network model to a bipartite entity-relation graph built from a KG. Recoin [2] suggests properties to an entity on Wikidata by collaboratively using the information about other entities that are similar to that entity; to ensure the high quality of the suggested properties, it sometimes involves much prior knowledge when defining the similarity between entities on Wikidata.…”

“…A closely related work to our instance completion problem is OKELE [7], which suggests relation-tail pairs to a long-tail head (i.e., a less popular/frequent entity) in a KG, using open Web data. Specifically, for a given head ℎ, OKELE first implements a relation prediction technique by applying an attention-based graph neural network model to a bipartite entity-relation graph built from a KG, predicting a list of relations; for each suggested relation , it then extracts and verifies potential tails from open Web data including semistructured vertical websites, unstructured plain text in Web content and structured HTML tables.…”

“…A straightforward solution to this problem is to combine the relation prediction and link prediction tasks [7]; for a given ℎ, we first predict a set of relations using a relation prediction technique, and then predict (ℎ, , ?) using a link prediction technique for each predicted relation .…”

Interpreting Open-Domain Modifiers: Decomposition of Wikipedia Categories into Disambiguated Property-Value Pairs

“…As it extracts semantic annotations over opendomain concepts (namely, over categories from Wikipedia), the proposed method falls under the area of open-domain information extraction (Ernst et al, 2018;Qu et al, 2018;Sun et al, 2018;Zhu et al, 2019;Zhan and Zhao, 2020;Dash et al, 2020;Cao et al, 2020). Previous work in that area often uses Wikipedia data (Tsurel et al, 2017;Konovalov et al, 2017;Korn et al, 2019;Bornemann et al, 2020).…”