Flexible Integration and Efficient Analysis of Multidimensional Datasets from the Web

Numeric data such as statistics and from sensors are increasingly published on the Web and – if brought together for comparisons and calculations – can answer important questions in science, industry, and politics. For instance, natural scientists compare rainfall values from sensors with hydrological estimations documented in a semantic wiki; financial analysts evaluate companies based on comparisons between KPIs from balance sheets filed with the SEC and daily stock market values from Yahoo! Finance; and citizens want to explore the GDP per Capita of different countries, independent from information sources such as Eurostat, the IMF, and the World Bank. However, the integration of datasets for analysis is difficult. First, heterogeneities remain a problem since publishers use different dimensions to describe numeric data, several identifiers for common entities, as well as differing levels of detail, units, and formulas. Second, aggregation and filtering operations, a varying selectivity of queries, and chains of joins over a growing number of possibly large datasets together with background information from the Web render analytical queries more complex than typical data analysis settings. The broad acceptance of the RDF Data Cube Vocabulary (QB) for publishing multidimensional datasets and of Online Analytical Processing (OLAP) interfaces for intuitive and interactive knowledge discovery call for a uniform view – the Global Cube – over available numeric data for exploratory analysis. This work presents four complementary contributions to query the global cube: 1. A mapping between the common model of data cubes and QB to use existing OLAP engines for efficient queries over datasets from the Web. 2. An algorithm to evaluate OLAP operations over data cubes using SPARQL queries over QB datasets for more flexible data integration based on RDF stores. 3. A method to optimise analytical query processing via materialised aggregate views in RDF, including an evaluation with a realistic benchmark. 4. A method to declaratively describe complex relationships between datasets for flexibly increasing the number of answers from the global cube. The contributions are applied to three scenarios in the areas of water resources management, company performance analysis, and Open Government Data exploration.