AI research tool: InsightPersona for more accurate political and media analysis

How often did certain politicians actually appear in video reports in the media during the election campaign and which excerpts or quotes made it into the media coverage? The Fraunhofer Institute for Digital Media Technology (IDMT) has developed a research tool with integrated AI models that can answer precisely such questions. It is called InsightPersona and makes it possible to search and analyze large media databases on individual people.

Did they really say that?

The Fraunhofer Institute presented a demo version of InsightPersona at Didacta 2025 in Stuttgart. The tool could be used in politics lessons for more detailed political analyses or in other subjects for the acquisition of media skills. The service is also aimed at publishers, research institutions, parliaments and political education organizations.

For example, InsightPersona can be used to show how often a candidate has appeared in the media in a certain period of time. To do this, the AI-based analysis software searches image and sound clips from an archive that you can create yourself. However, the tool also allows further analysis. Both the excitement of the voice and the topics mentioned in the excerpts are analyzed. The excitement is visualized by a colour scale, the topics by word clouds.

In order to be able to examine whether a particular quote has been shortened or taken out of context, for example, the excerpts are specified in detail and compared with each other. The archive used by the Fraunhofer Institute makes it possible to see, for example, which TV channel has shown longer or shorter excerpts or which excerpts – are more similar to each other and which are not. This also makes it possible to see whether a person was allowed to have their say or whether commentary predominated. This can also be analyzed in terms of gender.

This analysis can be supplemented by another feature that is still to be integrated into the dashboard. The Fraunhofer Institute has included publicly accessible speech transcripts from parliaments in the demo archive. With an integrated GPT 4.0, these transcripts can also be browsed in dialog with a chatbot. So if only very short excerpts from a debate were shown in media snippets, the speech transcripts can show the entire speech and list further statements on a specific topic from other transcripts. This could also be linked to longer video contributions, which can also be found in the parliamentary archives. This can make it clearer how a person actually presents a political issue or how the presentation has changed over various speeches. Various election programs could also be searched through more easily using this section.

Further development following a research collaboration

Researchers from both Oldenburg and Ilmenau are involved in the development of InsightPersona. Oldenburg is home to the Institute of Hearing, Speech and Audio Technology HSA, which conducts research into speech and event recognition, sound quality and speech intelligibility.
In Ilmenau, work is being carried out on AI-based recognition and classification of audio and video data, among other things.

The idea for the research tool came to the Fraunhofer employees during a collaboration with sociologists who were conducting a study on the question: "Climate change and biodiversity: What does television show – What do viewers want?"
The study was initiated by the MaLisa Foundation and the four television stations ARD, ZDF, ProSiebenSat.1 and RTL Deutschland. For the program analysis carried out in the context of the study, the Fraunhofer IDMT recorded EPG-controlled programs in a fixed period in 2022 and then analyzed them automatically.
A total of 1.4 million broadcast minutes were recorded, transcribed, analyzed and made available in a dashboard for further evaluation steps. According to Fraunhofer, this corresponds to around 2.7 years of broadcast material and 63 terabytes of program data.

Various licensing models are available for the InsightPersona software solution. According to Fraunhofer, it can be run locally by the customer (on-premises) or by a service provider (off-premises). The offer is individually customizable.

(kbe)