Knowledge Technologies

We research the automated generation of knowledge graphs and the integration of knowledge from heterogenous sources. Knowledge graphs of structured data and knowledge can be semantically fused and are often the reason why Artificial intelligence (AI) applications can be so easily explained and their results so easily understood.

Increasing digitalization means the quantity of data available worldwide doubles in size every two years – and with it grows its diversity. This explains the increasing trend for using knowledge graphs integrating many different data sources. This technology is the basis for many AI applications and AI assistants that would otherwise not work without them: Knowledge graphs are behind solutions used to retrieve information but also behind question answering systems such as those employed in chatbots.

We rely on uniform vocabulary and standard data format conventions, particularly from linked open data (W3C and Resource Description Framework, RDF) in order to generate automated global knowledge graphs. We worked with Bonn university to develop open source techniques to generate large knowledge graphs and retrieve information from them. We also contribute to many open knowledge graphs such as the DBpedia knowledge graph extracted from Wikipedia. Many of these are freely available.

Go to our publications

Research priorities

Our research is geared towards modeling terminologies, norms and domain-specific specialist knowledge from different areas as well as from standardized vocabularies. This means we can automatically and semantically link data pools with external sources. For example, a resulting knowledge graph created in this way can model a company’s entire knowledge base. They help create applications for searching for information or building dialog systems developed using company-specific knowledge.

Embedding knowledge graphs

Knowledge graphs can be integrated into AI architectures (e.g. deep neural networks) with the help of embeddings. This opens up a wide range of potential applications such as the prediction of as yet unknown facts using existing related facts.

Automated construction of knowledge graphs

We are conducting research to largely automate all steps involved in building and expanding knowledge graphs. We can, for example, extract knowledge graphs from different relational data structures (databases, spreadsheets, etc.), from JSON, XML files and, in some cases, from text sources.

Enterprise data integration

We integrate data with knowledge derived from heterogenous, distributed sources to build company-specific knowledge graphs. This integration is based on standardized vocabularies.

Highlights

Scalable Semantic Analytics Stack (SANSA Stack)

In cooperation with Bonn university and other research partners we have developed a library to help us process large knowledge graphs. The application is based on hybrid AI techniques and combines Machine Learning processes with semantic analysis methods. Large quantities of data can be processed in a highly scalable fashion and analyzed thanks to the open source software.

Technology for secure data spaces

Our knowledge technology is used in projects involving international data spaces sponsored by the BMBF (German Federal Ministry for Education and Research) and the Fraunhofer-Gesellschaft. This is what our scientists have created the reference architecture for and why they are developing an information model as well as domain-specific vocabularies.

Publication awards: Ten Year Best Paper Award

The “DBpedia: A Nucleus for a Web of Open Data” publication, which our Lead Scientist Prof. Jens Lehmann worked on with others, was awarded the Semantic Web Science Association’s (SWSA) SWSA Ten-Year Award in 2017. The awarding committee explained that it was chosen as the winner because it is still the most influential and most frequently cited paper on the Semantic Web despite it being ten years since its first publication.