Entity Alignment in Multilingual Knowledge Graphs

Abstract

Today there are billions of documents on the web. The amount of data we are generating each day has great value to be mined. All the challenges that cloud the aim to understand and use that data surround the lack of a human comparable inference and understanding capability. There has been many advances in NLP in the past decade but we have just started scratching the surface. Some of the biggest challenges are that information is highly unstructured. Knowledge graphs provide a way to structure information and also map out relations between the data. Knowledge graphs are a structured way to represent information in the form of fact tuples. The most common form being (subject, relation, object). Knowledge graphs have contributed to many nlp challenges including information retrieval, question answering etc. Prime example being search engines and AI personal assistants. Even though we have benefited from knowledge graphs a lot, they suffer with problems like sparse information representation, especially low resource languages. This has prompted research into knowledge graph completion techniques, knowledge graph alignments between very heterogeneous monolingual or multilingual knowledge graphs. In this report, we explore the ideas of Knowledge graph alignment via entity alignment techniques. The work that has lead the state of the art, the motivation and the novelties. We then look at some key architectural components which we think contribute to the gerneral improvement of Enitity alignment models. Finally we propose to develop a conglomerated model incorporating the best practices and state of the art techniques to build a robust, end to end entity alignment system.