Due to the growing fragmentation of the TV content offer, the role of super aggregator has gained importance in recent years: a combined offer of different autonomous TV services in a single unified service.
From a technical point of view, one of the main challenges is the unification of the offer: when the user searches for some specific content, it is essential that it is not presented with duplicates, even if the content is offered by different providers.
The unification of the offer is carried out at the metadata level, generally based on a semi-automatic model with a manual resolution of doubtful cases. Current algorithms to identify duplicate content are generally simple and robust in determining matches, but quite imprecise, which increases the manual work of reviewing non-duplicates.
The objective of TV Match projects is to improve existing and/or develop new AI-based algorithms and the core data processing architecture to:
- Automatically determine whether or not a new TV metadata entry refers to the same audiovisual content as a pre-existing entry stored in a global metadata repository.
- Automatically determine whether an image matches in a global image repository.
The technology developed in the TVMatch project will replace the current duplicate detection algorithm (both textual and graphical metadata) within the MediaStream information import subsystem. The KPI of merit in the integration is the significant decrease in the percentage of ingested metadata entries that require manual oversight by a human operator.