Active learning for entity alignment

Oshada Senaweera
Knowledge Graphs (KGs) represent real-world information or facts in the form of entities and relationships between them. In KGs, facts are represented in the form of "SPO" triples (subject, predicate, object). Popular examples of KGs are Wikidata, YAGO, DBpedia, NELL, Freebase and Google Knowledge Graph. Many of the KGs that have been created are done separately for a particular purpose. However, applications that use KGs have a more diverse knowledge requirement that a single KG cannot
more » ... To tackle this problem, it is essential to integrate multiple KGs into a unified KG, which can satisfy the diverse knowledge requirements of applications. Integration of these heterogeneous KGs can be done via entity alignment, i.e. identification of entities in different KGs that represent the same real-world entity. Manually doing so might ensure high quality, but soon becomes infeasible when confronted with large graphs. Hence, this study "evaluate different Active Learning Heuristics for the task of entity alignment in KGs" so that the best performing AL heuristics can be identified. To achieve this, instances(sample points) will be picked according to an active learning heuristic and the relationship between performance and the number of instances is studied. AL heuristics implemented in the study are centrality-based and model-based. As per the findings, best performing AL heuristics are based on the centrality where betweenness centrality performs the best.
doi:10.5282/ubm/epub.74084 fatcat:jjkkwjrcqvhvve2bfsmijwcdx4