A knowledge graph is a graph that stores semantic information in graph form. Each edge in the graph is a triple of the form (subject, predicate, object), also called a fact, indicating that two entities are connected by a specific relation, e.g., (GeorgeLucas, DirectorOf, StarWars). Large exemplary knowledge graphs are for instance Freebase (Bollacker et al 2008), Wikidata (Vrandecic and Krötzsch 2014), and DBpedia (Lehmann et al 2015).
Knowledge graph embeddings are a recent research direction in which the idea is to embed components of a knowledge graph into continuous vector spaces, so as to preserve the semantic meaning of components. These embeddings can then be used to benefit many tasks such as inferring new triples in the knowledge graph, for extracting new relations between known entities from text, for classifying entities, or for extracting new entities from text. A plethora of embedding approaches have been developed over the past few years, for a survey see Wang et al (2017).
One downside of current knowledge graph embedding techniques is that they typically assume that the whole knowledge graph must be known before training. This is problematic because large knowledge graphs receive updates frequently meaning their embeddings have to be retrained from scratch every time. As a result there have been at least two works that propose incremental knowledge graph embeddings, i.e., approaches that are able to update previously trained embeddings with new knowledge: Tay et al (2017) and Jia et al (2018).
The task of this thesis is to implement at least the two last mentioned papers and compare them for training large, frequently updated knowledge graphs along multiple dimensions. For a master thesis, an additional requirement is to develop an incremental variant of an existing knowledge graph embedding in cooperation with the thesis supervisors that is novel in some way.
The thesis can be written either in English or German.
Bollacker, K., Evans, C., Paritosh, P., Sturge, T., and Taylor, J. (2008). Freebase: A Collaboratively Created Graph Database For Structuring Human Knowledge. SIGMOD.
Vrandecic, D., and Krötzsch, M. (2014). Wikidata: A Free Collaborative Knowledge Base. Commun. ACM.
Lehmann, J., Isele, R., Jakob, M., Jentzsch, A., Kontokostas D., Mendes, P. N., Hellmann, S., Morsey, M., van Kleef, P., Auer, S., and Bizer, C. (2015). DBpedia – A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia. Semantic Web.
Tay, Y., Tuan, L. A., and Hui, S. C. (2017). Non-Parametric Estimation of Multiple Embeddings for Link Prediction on Dynamic Knowledge Graphs. AAAI.
Wang, Q., Mao, Z., Wang, B., and Guo, L. (2017). Knowledge Graph Embedding: A Survey of Approaches and Applications. IEEE Transactions on Knowledge and Data Engineering.
Jia, Y., Wang, Y., Jin, X., Lin, H., and Cheng, X. (2018). Knowledge Graph Embedding: A Locally and Temporally Adaptive Translation-Based Approach. TWEB.
Lukas Schmelzeisen, email@example.com, Room B104