Linked Open Data initiatives have made available a diversity of scientific collections where scientists have annotated entities in the datasets with controlled vocabulary terms from ontologies. Annotations encode scientific knowledge, which is captured in annotation datasets. Determining relatedness between annotated entities becomes a building block for pattern mining, e.g. identifying drug–drug relationships may depend on the similarity of the targets that interact with each drug. A diversity of similarity measures has been proposed in the literature to compute relatedness between a pair of entities. Each measure exploits some knowledge including the name, function, relationships with other entities, taxonomic neighborhood and semantic knowledge. We propose a novel general-purpose annotation similarity measure called ‘AnnSim’ that measures the relatedness between two entities based on the similarity of their annotations. We model AnnSim as a 1–1 maximum weight bipartite match and exploit properties of existing solvers to provide an efficient solution. We empirically study the performance of AnnSim on real-world datasets of drugs and disease associations from clinical trials and relationships between drugs and (genomic) targets. Using baselines that include a variety of measures, we identify where AnnSim can provide a deeper understanding of the semantics underlying the relatedness of a pair of entities or where it could lead to predicting new links or identifying potential novel patterns. Although AnnSim does not exploit knowledge or properties of a particular domain, its performance compares well with a variety of state-of-the-art domain-specific measures.