The approach
VRT’s media library contains millions of items and a variety of media types such as video, news articles, voicerecordings, and images. This content is spread over multiple database systems, each with its own unique way of storing metadata and substance. These libraries are maintained and expanded by experienced archivist teams but don’t have a single business-friendly search and exploration function across systems.
Inspired by the popularity of knowledgegraphs, VRT is interested in using the technology on top of their media archives to enable smart content search and exploration by archivists, journalists and program creators. Our members Dotdash and Nemeon understood the challenge and got to work with an AWS solution.
VRT
VRT is the national public-service broadcaster for the Flemish Community of Belgium. Among their channels are VRT 1, Canvas, Ketnet, Sporza and the online channels VRT NWS and VRT MAX.
The solution
On AWS, we built a knowledge graph proof-of-concept on top of various media libraries that connects different metadata types and content. This graph ingests and relates content by topics, entities, locations, contributors, and program structure. Natural language processing is used to extract important named entities (persons, organisations and locations) from content. Wikidata is used as external source to enrich and relate the named entities. Content and related topics are recommended by using the structure and context from the knowledge graph.
Firstly, we improved data exploration. We showed the possibility that after a google-like search, a visual representation of connected content gives an easy way to explore new and interesting content from thesame or related topics.
Secondly, we reduced support needed from archive experts. This means that in the future business users can explore media more autonomously, freeing up time for archive experts tofurther improve the data governance of media archives.
Finally, we made context aware search and recommendations possible. Currently locked potential included that more relevant content could be found and recommended for a given search term and content item. A search or recommendation for “Calatrava” could now include the train station of Luik, when that information was originally not in the data.
Benefits
Improved data exploration
We showed the possibility that after a google-like search, a visual representation of connected content gives an easy way to explore new and interesting content from the same or related topics.
Reduced support from archive experts
In the future business users can explore media more autonomously, freeing up time for archive experts to further improve the data governance of media archives.
Context aware search & recommendations
Currently locked potential includes that more relevant content can be found and recommended for a given search term and content item. A search or recommendation for “Calatrava” could now include the train station of Luik, when that information was originally not in the data. Furthermore, search results can filtered more efficiently based on context aware facets.