2018/12/07 by Bahare Fatemi, Fatemi, Bahare, Siamak Ravanbakhsh +3
Computer Science · #Advanced Graph Neural Networks #Computational Physics and Python Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1812.03235
openalex publication_date 2018/12/07 · openalex created_date 2022/08/01 · openalex updated_date 2026/07/28
Knowledge graphs are used to represent relational information in terms of\ntriples. To enable learning about domains, embedding models, such as tensor\nfactorization models, can be used to make predictions of new triples. Often\nthere is background taxonomic information (in terms of subclasses and\nsubproperties) that should also be taken into account. We show that existing\nfully expressive (a.k.a. universal) models cannot provably respect subclass and\nsubproperty information. We show that minimal modifications to an existing\nknowledge graph completion method enables injection of taxonomic information.\nMoreover, we prove that our model is fully expressive, assuming a lower-bound\non the size of the embeddings. Experimental results on public knowledge graphs\nshow that despite its simplicity our approach is surprisingly effective.\n