How machine learning can help solve the Big Data problem of video asset management

Abstract

This paper highlights similarities between problems in video asset management and Big Data management, and outlines how machine learning can assist in solving them. Machine learning has reached fever pitch in the industry as both the excitement around the technology, and the usefulness in addressing return on investment problems converge. As models are used to analyse video content for faces, emotions, objects, text, landmarks and more, digital asset management (DAM) systems must account for the influx of a tremendous amount of new metadata, often with varied structures and schemas. The value of this highly contextual metadata for every video asset an organisation owns is clear, but it is critical that any DAM system incorporating such new metadata does so in a way similar to how enterprises manage Big Data. This paper advises DAM owners and developers to consider using schema-less data stores to incorporate the data output from machine-learning models, and rethinking certain expectations regarding data management in order to get in front of a growing problem.