Mining & Harvesting – Text and Media: Impact at Cataloguing in the ORF-Archives
While Data-Mining is the darling in nearly all presentations about future archive work, Data-Harvesting is quite seldom mentioned and left the role of the ugly duckling. The short overview and presentation of both technology-approaches used in the ORF-Multimedia-Archives Department will try to demonstrate, that Harvesting is quite underestimated, while too high expectations are placed on Mining, whereby in practical use the promised savings can mostly not be achieved. On the other hand, it is also apparent that some worthwhile uses of mining are not recognised in purely theoretical planning and that new areas of application are only identified when they are used on a large scale; while the use of Harvesting can be planned and prepared very well theoretically. The presentation will try to shed some light on the pros and cons of both approaches, based on the experience gained during the last years in the ORF-Archives.
Stairway to heaven? AI at a public broadcaster, the case of the RTVE archive
After 3 years of testing AI solutions, in 2020 RTVE launched an AI tender to automatically catalog 11K hours of the video archive. It was a long and winding road before the tender was awarded and at that time we felt it was a great success for the archive. Our dreams have come true thanks to the decided support of the RTVE innovation area and the involvement of other technical areas. The archive has managed to lead a relevant project to ensure its future survival, but we soon discovered how long stairways to heaven can be.
Towards unsupervised content-based segmentation and description of TV programs using the CLIP model
The OpenAI model clip allows to generate embeddings to compare images and text in a common semantic vector space. In the framework of the RTVE University of Zaragoza Chair, several applications of this model have been explored for TV programs: from natural-language-based search to video segmentation and description.