Service-centric Networking


The goal of this research project is the conception and development of a holistic AI tool for the automated content production and knowledge extraction for technical and scientific texts such as scientific and medical journal articles. Since the documents that are used in this project are publications, there are no privacy issues to be adressed. Due to the unusual structure, specific terminology and high-complexity of these documents, human experts will contribute and provide their feedback in all critical steps of the process. In this way the reliability and accuracy of the results will be enhanced. The solutions that are currently available on the market:

  1. Black-box solution: In this case, the user does not know which methods will be used. This lack of transparency, can lead the automating existing processes to a lack of acceptance. Moreover, the offered solution is general purpose and it doesn’t work in individual use cases where more customised solutions are required.
  2. Data source: Only a very limited collection of data structures can be processed. In addition, different data sources are not aggregated and/or sufficiently pre-processed specifically for the system.
  3. Method selection: The methods used are limited to the simplest models and do not take into account the current developments in the field of machine learning. As a result, different approaches should be taken into consideration in order to find the most suitable one for each use case. Furthermore, human experts can provide their feedback to assess the efficiency of each method.

The goal of this research project is to address the above mentioned problems and to solve a specific use case in the field of content production, in the the form of a transparent holistic AIOps module with the essential contribution of human experts. For this purpose, innovative methods for the aggregation of different data sources are designed and a combination of machine learning methods and statistical methods with and without labelled data are evaluated. By preprocessing the data (documents) in a system-compatible manner and selecting complementary methods, specific knowledge from the data (documents) will be automatically extracted. For instance, text summarisation from scientific journal articles or extracting specific knowledge such as side effects from specific medicines based on research studies mentioned on the scientific journal articles. Human experts contribute with labelling the data for evaluating the results from the automated methods with choosing between different methods the one that suits better the data or by providing feedback to the system in order to be improved.


Period:04/ 2021 - 12/ 2022 
Team Members:Maria Mora MartinezAikaterini Katsarou
Partner:Holtzbrinck Publication Group  
Funding by:Software Campus, Federal Ministry of Education practical and Research (BMBF)