Open innovation: Geneea innovates by "geo-documenting" France Bleu Drôme Ardèche’s contents
Founded in 2014, Geneea used to be a startup, but has since they succeeded in growing on both its domestic market and internationally in just a few years. This Czech Republic scaleup specialized in natural language processing has worked for various European media. From writing newspaper articles automatically to recommending relevant photographs to illustrate an article, the scope of their software, Frida, is wide. In 8 years of existence, the scale-up has grown well. It went from 4 to 15 employees and now collaborates with companies in several European countries. As a member of the European network Future Media Hubs, Radio France offered Geneea to initiate a Sandbox project to accelerate their named entity recognition technology and allow the Radio France local station France Bleu Drôme Ardèche’s teams to produce maps of the places mentioned in four of its broadcasts.
What were the tests conducted with the local station France Bleu Drôme Ardèche?
Geneea analyzed 4 of the Radio France local station France Bleu Drôme Ardèche programs: Cap en Drôme-Ardèche, On cuisine ensemble, and the daily 8AM and 6PM news. Each episode of these programs was transcribed using the standard Amazon tools, that Geneea customized and refined for this experiment. The transcripts were then analyzed by Frida. Afterwards the information was combined with external data, such as the official INSEE geographic code, the population of Drôme and Ardèche, or the town of the listeners that called the local radio station. Thanks to this data and the results of the analyses, they produced several maps to show the places in Drôme, in Ardèche and in France most often mentioned during the programs. Not surprisingly, the most populated cities were also the most mentioned (Valence, Aubenas, Privas...). However, some areas were less mentioned in the regional news but were still the source of a significant number of calls from listeners. Beyond a better knowledge of France Bleu Drôme Ardèche’s audience, the maps helped the local editorial team implement more concrete actions using this data. In total, nearly 200 episodes were analyzed, and more than 3000 places were identified as named entities. Frida also highlighted other named entities within the transcripts produced: public figures, organizations, dates, events, etc.
What was the purpose of this Sandbox?
Taking part in Sandbox allowed Geneea to improve their software Frida and to automated it so that it produces daily results on its own. It also benefited from feedback given by Radio France’s teams involved in the experiment. Thus, Frida became better at correcting transcription errors and inaccuracies in French, a language that is difficult to grasp for natural language processing software. Radio France’s Sandbox program also allowed them to work on data visualization. Frida can now generate maps to show the results more clearly, an option that Geneea did not have before the experimentation.
Discover more innovative projects done by Radio France (website in French)
Visit the website of Geneea
