Effective entertainment discovery solutions require a deeper understanding of content, and one approach to harnessing this knowledge is extracting semantically-relevant metadata.

This paper explains how to use a combination of semantic graphs and machine learning (ML) to automatically generate structured data, recognise important entities/keywords and create weighted connections for more relevant search results and recommendations.

For example, the movie The Big Short can automatically produce entities, such as “hedge fund” and “subprime lending,” which are thematically relevant and therefore given a high weight. By inferring relevant entities through these underlying technologies, metadata results are richer and more meaningful, enabling faster decision-making for the consumer and stronger viewership for the content owner.


Today’s consumers have the advantage of choice – but from an ocean of content, including movies, programmes, news and short-form video from an array of linear and streaming services. Because there is so much content, largely lacking structured metadata, viewers are frustrated – they can’t find what they want to watch quickly and easily.

Moreover, a 2016 consumer research study by ‘TiVo’ identified a phenomenon called “show-dumping,” where consumers simply give up on programmes due to the challenges involved in accessing them. Show-dumping leaves content owners with a big problem: they heavily invest in producing excellent content, yet struggle to ensure consumers can find it.

A deeper understanding of content is required to create intelligent solutions that can overcome the challenges faced by consumers and content owners alike. Using traditional statistics-driven models for entity extraction will not solve the problem, as they lack semantic understanding. Combining machine-learning methods and semantic graphs is a unique way to add much-needed context and can alleviate consumer frustration, as well as strengthen viewership for content owners.

Historically, semantic graphs have helped a great deal in question-answering ‘Dali et al’ and text summarisation ‘Moawrd and Ared’. In this paper, we delve into ways to leverage the importance of the nodes in a semantic graph to train a machine-learning model that will automatically determine the relevance of an entity in a given blurb of text, thus serving up better results for consumers to find what they want to watch.

Download the full technical paper below