Computational methods and optimizations for containment and complementarity in web data cubes

•Definitions of full containment, partial containment and complementarity for RDF data cubes.•Presentation of baseline quadratic method for computation of the defined relationships.•Presentation of three alternative methods for efficient computation of containment and complementarity relationships.•Experimental evaluation of efficiency and scalability on real world and synthetic data. The increasing availability of diverse multidimensional data on the web has led to the creation and adoption of common vocabularies and practices that facilitate sharing, aggregating and reusing data from remote origins. One prominent example in the Web of Data is the RDF Data Cube vocabulary, which has recently attracted great attention from the industrial, government and academic sectors as the de facto representational model for publishing open multidimensional data. As a result, different datasets share terms from common code lists and hierarchies, this way creating an implicit relatedness between independent sources. Identifying and analyzing relationships between disparate data sources is a major prerequisite for enabling traditional business analytics at the web scale. However, discovery of instance-level relationships between datasets becomes a computationally costly procedure, as typically all pairs of records must be compared. In this paper, we define three types of relationships between multidimensional observations, namely full containment, partial containment and complementarity, and we propose four methods for efficient and scalable computation of these relationships. We conduct an extensive experimental evaluation over both real and synthetic datasets, comparing with traditional query-based and inference-based alternatives, and we show how our methods provide efficient and scalable solutions.