JoIE II - Journalistic Information Extraction

In the Journalistic Information Extraction (JoIE) project, the problem of extracting information from unstructured sources, which is relevant to (data) journalism, was therefore addressed in an initial research and development phase. An accompanying doctoral project showed that systems for processing and indexing information are of crucial importance for the journalistic user group. Tables in particular were identified as a promising resource for journalistic work, as they have a high information density and are found in many scientific documents. However, there is a lack of tools for extracting information from tables that are suitable for users without specialised technical knowledge. It was also possible to identify the shortcomings of current state-of-the-art systems to support journalists. The main obstacles to using these systems are usability, the high level of specialised technical knowledge required of the user and the lack of explainability and user interaction with the underlying complex models.
In the second phase of the project, new interactive systems are to be developed using a combination of explainable AI, data programming and active learning methods, which will also enable users without technical expertise to gain knowledge from large amounts of information. The aim of JoIE remains to support science journalists focussing on data in their research work. In contrast to existing approaches or partial solutions, we are focussing on an approach tailored to science journalists that is based on trustworthy and comprehensible procedures, allows user feedback and is available to the community as open-source software in accordance with the principles of open science.