While traditional machine learning models often constitute black boxes whose predictions are hardly comprehensible by humans, white box models make their predictions in a transparent way. Such white-box models are particularly promising to apply to knowledge graphs which represent knowledge in a human-readable form, e.g., as subject-predicate-object triples such as ("Paderborn", "has major", "Michael Dreier") or ("Paderborn", "has population", "151,633"). Popular examples of knowledge graphs include DBpedia, YAGO, and Wikipedia and they are heavily used by search engines such as Google, Bing, and Yahoo. While first approaches for learning understandable rules have been proposed (e.g., EvoLearner, see below), they do not scale to real-world knowledge graphs with millions of triples yet. The goal of this project group is to design novel algorithms for learning rules from positive and negative examples in knowledge bases in a scalable way.
Heindorf, S., Blübaum, L., Düsterhus, N., Werner, T., Golani, V. N., Demir, C., and Ngomo, A. C. N. (2022). EvoLearner: Learning Description Logics with Evolutionary Algorithms. Accepted at WWW'22. arXiv preprint arXiv:2111.04879. https://arxiv.org/pdf/2111.04879.pdf
L.079.07006 Project Group: Explainable Artificial Intelligence II (in English)