Representing Evolving Knowledge Graphs through Incremental Embeddings[go to overview]
Knowledge Graph Embedding describes techniques that learn numerical representations for a knowledge graph’s entities and relations. Typically knowledge graphs evolve over time. However, most such methods are mainly designed for knowledge graphs whose set of facts is assumed to be unchanging. If a graph is manipulated, these methods have to reject their learned representations and learn them from scratch. The research area of Incremental Knowledge Graph Embedding tackles this issue and describes methods that are able to adapt representations incrementally after a graph modification. Up to now, there is no benchmark in the literature to assess such techniques appropriately. For this reason, I developed a evaluation framework in my thesis and compiled a first benchmark dataset called WikidataEvolve which I want to share with you in this presentation.
29.10.20 - 10:15
via Big Blue Button