Link prediction (LP) aims to tackle the challenge of predicting newfacts by reasoning over a knowledge graph (KG). Different machinelearning architectures have been proposed to solve the task of LP,several of them competing for better performance on a few de-factobenchmarks. The problem of this thesis is the characterization of LPdatasets regarding their structural bias properties and their effects onattained performance results. We provide a domain-agnostic frameworkthat assesses the network topology, test leakage bias and sampleselection bias in LP datasets. The framework includes SPARQL queriesthat can be reused in the explorative data analysis of KGs foruncovering unusual patterns. We finally apply our framework tocharacterize 7 common benchmarks used for assessing the task of LP. Inconducted experiments, we use a trained TransE model to show how the twobias types affect prediction results.Our analysis shows problematic patterns in most of the benchmarkdatasets. Especially critical are the findings regarding thestate-of-the-art benchmarks FB15k-237, WN18RR and YAGO3-10.