Extraction and utilization of facts from medical publications
Initial situation
In the medical field, there are a large number of specialist scientific publications in which new findings from research are constantly being published. The volume of knowledge is almost unmanageable, even for experts. Thieme therefore provides an expert audience with easy-to-understand, short summaries of medical and scientific research results. The task for the Quick Check was to evaluate the possible applications and benefits of AI in the creation of such summaries.
Solution idea
Based on a common understanding of the task and the associated challenges, an approach was developed that comprises the following sub-steps:
-Extraction of text passages
-Summary to full text
-Rewording of the text
-translation
The NLP framework SpaCy was used to extract the text passages. The reformulation of the text passages found and the translation from English into German was carried out using AI language models from the Hugging Face platform.

Benefit
The quick check showed that all subtasks can be implemented. The use of large language models with transformer architecture makes it possible to extract text passages with less effort and therefore better efficiency. The first useful results can be achieved with just a few training and test examples. The solution approach developed for the medical field can later be transferred to other subject areas.
Implementation of the AI application
In the Quick Check, the entire procedure was implemented as an example for the "study participant" feature using scientific studies from specific subject areas. By integrating the functions into Fraunhofer IAO's text comprehension platform Aikido, an initial prototype was realized that makes the results of the AI available in the form of an assistance function. The assistance solution makes it possible to collect user feedback later on during use, allowing further training data to be obtained, which improves the quality of the AI suggestions.
