About

As ITV’s streaming catalogue has expanded, traditional genre-based metadata has become increasingly limiting. With ITVX now offering tens of thousands of hours of content, improving discovery and personalisation requires a deeper understanding of what programmes are about.

Following on from ITV’s previous work presented at Data Decoded Manchester, this session explores how those foundations have been extended using modern AI techniques. George and Alfie will share how pre-trained transformer models (such as sentence embedding and Named Entity Recognition models) are used to enrich content metadata at scale, helping surface themes, topics and relationships across the catalogue, and drive value across a wide variety of use cases at ITV.

Takeaways

• How pre-trained transformer models can be applied pragmatically to enrich content metadata
• Why we shouldn’t throw everything at LLMs – and how the techniques that underpin them can be used more effectively for repeatable, production-grade tasks
• Why enriched metadata is vitally important in delivering value in the entertainment industry