Hasso-Plattner-Institut
Prof. Dr. Felix Naumann
  
 

24.02.2021

Paper accepted at ESWC 2021 conference

Nitisha Jain, Jan-Christoph Kalo, Wolf-Tilo Balke, Ralf Krestel

We are happy to announce that our full paper titled "Do Embeddings Actually Capture Knowledge Graph Semantics?" has been accepted at the research track of the Extended Semantic Web Conference 2021 (ESWC 2021). This paper reports our findings as negative results when employing knowledge graph embeddings for semantic representation of facts. 

This work is the result of a successful research cooperation with Jan-Christoph Kalo and Wolf-Tilo Balke from Technische Universität Braunschweig.

 

Abstract

Knowledge graph embeddings that generate vector space representations of knowledge graph triples, have gained considerable popularity in past years. Several embedding models have been proposed that achieve state-of-the-art performance for the task of triple completion in knowledge graphs. Relying on the presumed semantic capabilities of the learned embeddings, they have been leveraged for various other tasks such as entity typing, rule mining and conceptual clustering. However, a critical analysis of the utility as well as limitations of these embeddings for semantic representation of the underlying entities and relations has not been performed by previous work. In this paper, we performed a systematic evaluation of popular knowledge graph embedding models to obtain a better understanding of their semantic capabilities as compared to a non-embedding based approach. Our analysis brings attention to the fact that semantic representation in the knowledge graph embeddings is not universal, but restricted to a subset of the entities based on dataset characteristics. We provide further insights into the reasons for this behavior. The results of our experiments indicate that careful analysis of benefits of the embeddings needs to be performed when employing them for semantic tasks.