Business Intelligence on Non-Conventional Data

The revolution in digital communications witnessed over the last decade had a significant impact on the world of Business Intelligence (BI). In the big data era, the amount and diversity of data that can be collected and analyzed for the decision-making process transcends the restricted and structured set of internal data that BI systems are conventionally limited to. This thesis investigates the unique challenges imposed by three specific categories of non-conventional data: social data, linked data and schemaless data. Social data comprises the user-generated contents published through websites and social media, which can provide a fresh and timely perception about people’s tastes and opinions. In Social BI (SBI), the analysis focuses on topics, meant as specific concepts of interest within the subject area. In this context, this thesis proposes meta-star, an alternative strategy to the traditional star-schema for modeling hierarchies of topics to enable OLAP analyses. The thesis also presents an architectural framework of a real SBI project and a cross-disciplinary benchmark for SBI. Linked data employ the Resource Description Framework (RDF) to provide a public network of interlinked, structured, cross-domain knowledge. In this context, this thesis proposes an interactive and collaborative approach to build aggregation hierarchies from linked data. Schemaless data refers to the storage of data in NoSQL databases that do not force a predefined schema, but let database instances embed their own local schemata. In this context, this thesis proposes an approach to determine the schema profile of a document-based database; the goal is to facilitate users in a schema-on-read analysis process by understanding the rules that drove the usage of the different schemata. A final and complementary contribution of this thesis is an innovative technique in the field of recommendation systems to overcome user disorientation in the analysis of a large and heterogeneous wealth of data.