Smart stacking of machine learning and semantic technologies in media management

Abstract

The television sector has well-established industry standards and conventions around the description of franchises, shows, episodes and other aspects of production and distribution. However, time-based metadata describing specific moments and clips are much rarer. This is in large part because, without augmentative or assistive technology, such metadata is hard to produce at scale and at sufficient quality. This paper describes a project at the US-based Public Broadcasting Service (PBS) that combined machine learning and semantic technology to provide time-based metadata describing moving image content that both linked back to and enriched the organisation’s broader information domain. The content chosen for the project was a selection of episodes from ‘Sesame Street’, the long-running and much-fêted show aimed at preschoolers produced by the Sesame Workshop, and the featured taxonomy (part of a larger in-development PBS knowledge graph) was developed by the PBS Education division.