Knowledge Acquisition Team Seminar (Talk by Rumana Ferdous Munne).

This is an online seminar. Registration is required.
【 Knowledge Acquisition Team】
【Date】2023/Juy/5(Wed) 11:00-12:00(JST)
*【Speaker】Rumana Ferdous Munne *

Title:
Entity Alignment and Attribute Enhancement between Knowledge Graphs

Abstract:
A Knowledge Graph (KG) is a knowledge model containing facts about real-world entities represented as a graph. It is a collection of interlinked descriptions of entities, relationships, concepts, and events. We have witnessed rapid growth in knowledge graph creation and application in the last few years. Several efforts have been made to develop knowledge graphs in general and specific domains such as DBpedia, YAGO, LinkedGeoData, and Wikidata and they have been served several fields of real-world applications from semantic parsing and named entity disambiguation to information extraction and question answering. These knowledge graphs contain millions of facts about entities. However, these knowledge graphs are far from complete and mandate continuous enrichment and enhancement.

One possible approach to enhance KG is integrating knowledge from various knowledge graphs based on their aligned information. In our research, we develop new effective methods to find aligned entities from different KGs first (entity alignment) and later enrich the KGs by enhancing their attributes. The task of entity alignment is to find entities in two heterogeneous knowledge graphs that represent the same real-world entity. Many knowledge graphs have been created separately for certain purposes with overlapping entity coverage. These knowledge graphs are complementary to each other in terms of completeness. Unfortunately, only a fraction of the entities stored in different KGs are aligned.

In this talk, we will discuss an embedding-based entity alignment method that finds entity alignment by estimating the similarities between entity embeddings. Then we will introduce a KG enhancement framework that enhances the attributes of Knowledge Graphs (KGs) by integrating multiple KGs that share aligned information. This integration process results in an enriched KG with improved attributes which leads to a more robust and complete knowledge graph.