[Back]

L. Wu, E. Sallinger, E. Sherkhonov, S. Vahdati, G. Gottlob:
"An Evolutionary Algorithm for Rule Learning over Knowledge Graphs";
Talk: KR4L 2020 - International Workshop on Knowledge Representation and Representation Learning co-located with the 24th European Conference on Artificial Intelligence (ECAI 2020, Santiago de Compostela, Spain; 2020-08-29 - 2020-09-08; in: "Proceedings of the International Workshop on Knowledge Representation and Representation Learning co-located with the 24th European Conference on Artificial Intelligence {(ECAI} 2020), Virtual Event, September, 2020", (2020), 52 - 59.

Logical rules allow us to declaratively encode expert knowledge,
express patterns and infer new knowledge from existing one in
Knowledge Graphs. However, the construction of rules is often a costly
manual process. In this work, we address the problem of learning rules
from the already exiting knowledge. Most of the existing rule learning
algorithms for Knowledge Graphs are based on inefficient full search plus
greedy pruning strategies that cover limited search space by considering
rules of a predetermined shape. We propose a learning algorithm where
a set of rules is learned via the process inspired by biological evolution
and that is geared towards optimizing a fitness function. Such evolutionary
algorithms typically cover larger search space efficiently and provide
multiple near-optimal solutions. We evaluate the proposed algorithm on
a number of public Knowledge Graphs and compare it to other rule
learning algorithms.

Rule Learning · Knowledge Graphs · Evolutionary Algorithms · Genetic Programming

Project Head Reinhard Pichler:
KnowledgeGraph