LSP.6. Model-Flexible Data Processing

Traditionally, analytics were all built around the multidimensional data model. Any kind of data was transformed to fit into a carefully design multidimensional database schema. Today, with the spreading of advanced analytical methods and more and more non-relational databases around, purely multidimensional data analytics infrastructure is dated and inapplicable in many application areas. For instance, social graph data is of increasing importance in analytics. Many advanced analytical methods originate from statistics and build on matrices. Both, graph data and matrices, cannot be naturally represented in the multidimensional data model. Data management solutions have to be prepared and need to tackle this challenge. Model-flexible data processing aims at building a data analytics infrastructure that is flexible towards the used data model. A more generic but adaptable data representation and a universal set of operations allow easily implementing data model specific databases and analytic tasks on top of the model-flexible analytics infrastructure. Importantly, it has to be easy to combine data from different data models for analytic tasks. Within this topic, different levels of model abstraction have to identified and analysed. Particularly, the generality and the operational power of each level have to be compared to each other to find a reasonable and useful balance between flexibility and operational power. While flexible regarding the data model, the analytics infrastructure should also provide a powerful set of data operations commonly used in analytics. Naturally, efficiency is the key in data management. Hence, it is essential to investigate efficient implementation of data representation and data operations for the chosen data model abstraction. Implementation concepts have to be experimentally investigated and evaluated.

Main Advisor at Technische Universität Dresden (TUD)
Co-advisor at Université Libre de Bruxelles (ULB)