Natural Language Understanding Wiki

The research on "Type-constrained Representation Learning in Knowledge Graphs" [1] aims to tackle the issue of graph scarcity and enhance the link prediction task in knowledge graphs (KGs).

Previous research has focused on building latent variable models. This research is an attempt to improve the semantics of the outcome of these models, through constraining the predicted links with knowledge on the relation types. These relation types define the semantic roles of the relations and can be either explicitly defined in the KG or inferred as domains and ranges based on example data. This inference relies on a CWA (Closed World Assumption): the semantics of the observed triples can be described through the observed triples.

The authors perform evaluation by applying prior type knowledge on top of latent models (RESCAL[2], TransE[3] and Knowledge Vault[4]) and observe improvements on three different datasets.

[1] D. Krompaß, S. Baier and V. Tresp. Type-constrained representation learning in knowledge graphs. In The Semantic Web–ISWC 2015, pages 640-655. Springer International Publishing, 2015

[2] M. Nickel, V. Tresp, and H. Kriegel. A three-way model for collective learning on multi-relational data. In ICML, pages 809–816. ACM, 2011.

[3] A. Bordes, N. Usunier, A. Garca-Durn, J. Weston, and O. Yakhnenko. Translating embeddings for modeling multi-relational data. In NIPS, pages 2787–2795, 2013.

[4] X. Dong, E. Gabrilovich, G. Heitz, W. Horn, N. Lao, K. Murphy, T. Strohmann, S. Sun, and W. Zhang. Knowledge vault: A web-scale approach to probabilistic knowledge fusion. In SIGKDD, pages 601–610. ACM, 2014.