As part of the agriProKnow project with industry partners I was involved in the implementation of an active semantic data warehouse for precision dairy farming. The project goal was to integrate relevant data from various sources, from sensors to internal and external databases, in order to detect signs of animal illness in the data before an acute outbreak of the illness. When specific events occur, e.g., the detection of signs of animal illness in the data, the data warehouse system should automatically execute the appropriate action, e.g., calling a veterinarian. An important aspect was data cleaning, especially handling missing data, and finding a data model that allows for comparison between farms.
The agriProKnow project successfully brought together agricultural engineers, sensor developers, veterinarians, statisticians, and business intelligence (BI) experts. As BI experts and data engineers, our group's role was central to the development of the data warehouse: We had to ensure that engineers, domain experts, and business people spoke the same language and agreed upon a common system architecture and data model.
In the agriProKnow project, we also developed a pattern-based approach to multidimensional data analysis where OLAP patterns represent frequently occurring, generic analytical queries. OLAP patterns helped cope with uncertain requirements regarding the involved veterinarians' information needs. The concept of OLAP patterns draws from my earlier research on reference modeling for data analysis. In this context, I have also participated in research on the concept of analysis graphs, which allow for the proactive modeling of interesting processes in multidimensional data analysis for documentation and communication purposes. The concepts of OLAP patterns and analysis graphs, respectively, are currently turned into doctoral theses, which I am co-advisor of.
Together with colleagues from the Fondazione Bruno Kessler in Trento, Italy, I am working on Knowledge Graph OLAP (KG-OLAP), a framework for representing, managing, and querying contextualized knowledge graphs. Hence, in KG-OLAP, knowledge facts are associated with points in a multidimensional space akin to the concept of OLAP cube, each point representing a context of applicability for the associated knowledge facts. The contexts are hierarchically organized; knowledge from more general contexts propagate to the more specific contexts. The KG-OLAP framework provides contextual operations and graph operations. Contextual operations allow for the selection and combination of knowledge facts from different contexts. Graph operations replace individual entities within a context with more general entities.
Related to KG-OLAP is the idea to employ business model ontologies in OLAP cubes for the representation of complex business facts for analytical purposes. Many real-world facts do not easily boil down to simple numeric measures. Business model ontologies may be used to represent complex strategic analyses of an organization’s situation and business environments. For example, a variant of iStar can be used to represent the analysis of an organization’s environment using the PESTEL framework. The REA framework can be used to represent value chains. Using RDF and ontologies opens such analyses up to more formal analysis.
Over the years, I have been involved in various projects with industry partners regarding the use of intelligent applications for air traffic management. In general, those projects aimed at leveraging semantic technologies, ontologies, and deduction systems for air traffic management. Frequent collaboration partners are FREQUENTIS and EUROCONTROL. We have even collaborated with NASA's Intelligent Systems Division, investigating differences in ontologies for air traffic management. These days I'm working on the SlotMachine project, developing a privacy-preserving online marketplace for slots in air traffic management.
In the BEST project with industry partners I took the lead of the work package that proposed semantic containers for packaging information in air traffic management. A semantic container comprises messages and other items that fulfill a membership condition, which is a semantic label that includes information about data provenance as well as geographic and temporal scope of the data in the container. The concept of semantic container has meanwhile been extended into a general-purpose data exchange framework by OwnYourData. The MyPCH project later employed semantic containers for the management of diabetes data. I was involved in that project in a consulting role regarding the inclusion of digital watermarking in semantic containers; results were presented at the 19th International Workshop on Digital Forensics and Watermarking.
Extending the concept of semantic containers, ATM information cubes allow for the hierarchical organization of semantic containers along different dimensions as well as the subsequent selection of relevant containers and abstraction of data items for specific tasks.
In the SlotMachine project, we developed a data sharing platform and marketplace for airlines to be able to optimize flight sequences in case of reduced capacity in the air traffic network. Airlines submit data about their preferences regarding the optimization, which allows to find a globally optimal solution to the flight prioritization problem. Airlines are compensated for giving up favourable slots, receiving credits that can be spent to prioritize flights in the future. The SlotMachine system employs evolutionary algorithms in connection with privacy-preserving computation to protect the airlines' private inputs while still allowing for optimization. The proposed distributed architecture for privacy-preserving optimization, although originally developed for the SlotMachine project, could also be potentially applied in other domains.