Broadband TV News speaks to Comigo COO Amir Eilat. Based in Israel and the United States, Comigo uses intelligence-based algorithms to enable TV service and content providers to create new revenue streams from a contextual services.
Broadband TV News: How much understanding is there of metadata in the broadcast industry?
Amir Eilat: There is a growing understanding in the industry of the importance of rich metadata and its influence on a good UX/UI.
The popularity of Netflix, Apple TV, Amazon Fire TV and alike, set a high bar in terms of creating a powerful metadata experience that accompanies user journey and content consumption. We see more and more leading providers in the industry designing a rich metadata UX. However, what looks good on the concept board and in presentations lacks in real life. This is mainly due to the fact that a lot of metadata is missing – so instead of beautiful imagery and text, we get a UI with a lot of ‘black holes’.
The only way to tackle this challenge is by taking the basic metadata, which currently exists and automatically enrich it with an unparalleled number of images, cast, bios, point of interest, etc. Using AI-based technology we’re able to build a vast knowledge graph from the content, where everything is an entity – actors, point of interest, sports players, events, etc, result in entities interweaving relentlessly, in turn, made available to the users empowering the OTT service’s immersion.
BTN: Does this vary between the broadcast and distribution sectors?
AE: One of the major differences between distributors and broadcasters used to be the relationship with the end-users. In the world of OTT, this difference is quickly becoming irrelevant as broadcasters and content owners are launching their own OTT services enabling them to have a direct relationship with their viewership. In this respect, the needs of both sectors are quickly becoming one. One could argue that broadcasters may want to further promote their content by adding a layer of commerce related to their content – deep learning can seamlessly connect online retailers catalogues contextually related – thus, facilitating the additional revenue stream on top of the standard OTT revenues.”
BTN: How does the Comigo approach differ to what’s already in the market?
AE: Unlike the traditional metadata providers which are providing static metadata, at the foundation of TV AI there’s the Knowledge Graph – a web of inter-connected entities that relate to the content and are induced by deep learning for Natural Language processing.
After an entity is created, it undergoes an enrichment process which finds relevant data from various metadata sources. The enrichment process scope ranges from complementing missing basic metadata (for example, in case of a content item with nothing but title it will find plot, cast, thumbnails etc), to find related metadata from various unstructured and semi-structured data-sources such as Facebook, Twitter, Instagram (for example, find related tweets to specific actor, or the actor’s Instagram account) all being done automatically by using machine learning.
In addition, a micro service architecture caters for the augmentation of entities from other domains that we believe to have added value when considering content metadata. For example, integrating sports real-time statistics, as a case-in-point: soccer, into our very own ‘Experience Intelligence Cloud’ in such a way that soccer-metadata (game events, player stats, team stats etc) are readily available as part of the actionable services exposed with APIs.
The relationships between the entities are also automatic – a football match is connected to the content entity in which the match is aired, and a footballer is a person entity that is linked to the match but also linked to a talk-show or a documentary that the player appears in. The beauty is the ability to fully expose the OTT service’s entire content in a way not possible before, emphasising to the end-users the value of the offering as a whole.
BTN: Is there enough room for consumer personalisation beyond what the operator might want to promote?
AE: Consumer personalization is done on-the-fly during a consumer’s journey through the different entities across the Knowledge Graph. Each user will have his own journey which will differ from any other user and fit his ever-changing interests and needs. We believe it opens the gate to a personalize ‘content cruising’ experience between VOD, EPG, sports, music and more – through the metadata which relies on common entities and the links between them – allowing the user to discover more content, consume more content and thus be even more satisfied from the OTT experience they are getting.
BTN: Can some systems be too clever for their own good and forget usability?
AE: The end-users keep evolving in the way they interact with apps and content. They expect their time to be optimised, they expect to be delighted by the services they pay for or to extract value from their ad exposure. We understand the challenges in introducing new features and functionalities to an existing OTT service. However, no one can afford to stand still, UX should grow in form and functionality, yet fit both the young ‘feature hungry’ and the older more conservative. TV AI accelerates the OTT attributes as well as allowing for each territory its own set of services fitting the local tastes and languages.
BTN: We’re familiar with broadcaster’s providing stats during major sporting matches. Can metadata be used to bring this to the consumer directly?
AE: Of course. We believe that sports and news are the main reasons people still watch live broadcast. EI Cloud’s ability to connect to multiple online data sources in order to bring sports events, players, stats and metadata enable enrichment of the viewing experience to new levels. Such that allow the users to remain within the TV experience in order to get all the information they wish to have and enable participation, cheering, following, betting, playing and more.
BTN: How can we make sure that the information going into the consumer’s home is accurate?
AE: One of the main objectives of creating a successful TV AI implementation is accuracy – identifying as much of the content with no false positive/false identifications. When we train our AI product, we employ comprehensive methods that handle the disambiguation of content items and the respectively derived data, crossing it with the entity based Knowledge Graph in order to achieve the most precise results. Once it passes the above processes it will use the profanity detection mechanism to make sure all the content is verified and ready for a public wide audience.