Skip to main content

Metadata Distillery and ExpLoitation System

Periodic Reporting for period 2 - MoDELS (Metadata Distillery and ExpLoitation System)

Reporting period: 2019-08-01 to 2020-07-31

The discoverability of high-quality content is only becoming more important. For video content producers, broadcasters and service providers to be successful in the attention economy, it is of utmost importance to optimise discoverability and meet the user experiences the viewers expect, to prevent them from going elsewhere. The current state of video metadata technology does not help. Although the production of video content is becoming more and more advanced; the metadata available for broadcast video often still resembles the quality and details as was provided in the past century. Much like the early search engines for web pages in the nineties, current video metadata engines and recommender systems lack an understanding of what video content is actually about. There are no smart algorithms. 90% of the metadata that is used as input only consists of whatever was manually added by the video producer or manually added afterwards by a third party.
The solution that Media Distillery presents is smart content recognition technology that automatically generates in-depth, descriptive information about video content in real time.
Especially at broadcasters and tv operators, hardly any of the available software is currently being used. These organisations don’t have the IT capacity and technical mind-set to make the transition towards the optimal user experience that the public demands. Due to this lack of IT knowledge, vast amounts of quality content is currently only used to fill archives. Millions of euros invested in (subsidised) TV content is unable to find for the interested audience. Strong online and often US-based competitors such as Netflix and YouTube are currently filling this gap as they have more funding and economics of scale.

Our main goal is to scale our products for adoption by broadcasters and TV operators, but also to enhance every analysis component in use by Media Distillery. To fulfil this condition, the engine requires algorithms that seamlessly adjust to new settings. In pursuit of this objective we will:
• Enhance our distillery system by integrating two new AI analysis components
• Refine current AI distillery algorithms (analysis components) with the latest insights
• Unlock services for users in by means of Multi-modal analysis integration.
• Data pre-processing, machine learning tasks, calculation and validation of scores for gathered data.
• Unlock services for the TV operators by means of media application for demonstration activities.
• Optimize components to reduce implementation times and costs.

Demonstrate the effectiveness of Media Distillery in four media applications
MoDELS will be integrated within four media applications at broadcasters and TV operators throughout Europe. MoDELS will be used to power recommendations & user search, support editorial content and optimize the user experience in broadcaster’s platforms. Within the context of these demo’s we will:
• Establish the requirements for integration of our data in broadcasting platforms.
• Incorporate user and application specific adaptations in our APIs / software.
• Establish the impact of the technology (financial / non-financial) on the users’ business.
• Generate data to optimize the developed technology.

Success in commercialisation strategy of our new business plan Implement the presented marketing and commercialisation strategy. After testing and validation, we aim to commercialize MoDELS across the broadcasting value chain. To be successful we need to focus on:
• Dissemination: attract business partners and customers to create market demand.
• Business Plan and commercialisation strategy: incorporating a detailed commercialisation strategy (dissemination and exploitation) and a financial plan in view of market launch.
• Establish agreements with key customers in all relevant domains within European broadcasters and distributors.
In the first half of our project we have contributed the following to the MoDELS objectives:
Enhance our distillery system by integrating two new AI analysis components
We have developed two new AI analysis components from scratch, both supporting the general objective of our project. The first component is the Automatic Cue Classifier, which can automatically detect channel logos and other cues that are important for EPG Correction product. The second component is the Overlay Text Recognizer, which can detect and read text that has been digitally added to videos. For instance, name tags in news programs. This helps in the discoverability and understanding of what content is about.

Refine current AI distillery algorithms (analysis components) with the latest insights
As part of the refinement of the existing AI algorithms we focused on improving the face recognition system, improving our languages models and bumping up the quality of our Optical Character Recognition (OCR) algorithm that we use for reading open captions (burned-in subtitles).

Optimize components to reduce implementation times and costs.
A major milestone we have achieved is that we manage to transform our Program Event Service to a scalable and easy to deploy service by applying modern technologies and state of the art best practices. Encouraged by this success we will continue to work on making the Program Event Service even more scalable and deployments even easier, such that we can provide our services to more customers in less time.

Demonstrate the effectiveness of Media Distillery in four media applications
As part of the goal to demonstrate our technology we have worked with two large European TV operators to perform a proof of concept project where we applied both new and improved AI components for one of our emerging products and tested with customers.
Success in commercialisation strategy of our new business plan Implement the presented marketing and commercialisation strategy.
As part of our commercialisation strategy we have executed extensive market research on the TV industry market in Europe and identified our prospects. Based on that research we sponsored and presented at numerous industry events, where we share our findings and insights about how AI can help broadcasters and TV operators.

In the end our efforts lead to a good amount of customers for our program event services and other products we offer, and leads to recognition of our contributions to the European media industry.
Our expected results until the end of the project are:
• We have launched two new products which are in use by broadcasters or tv operators that help them to improve the user experience of their video services
• We have validated two new emerging products with paying customers.
• Our products are impacting tens of millions of video consumers on a daily basis, indirectly by implementing our technology at tens of TV Operators and broadcasters

Potential impact:
• Improve the discoverability of content created and/or provided by European media service providers
• Improve the user experience by removing user annoyances and thus prevent churn to other US-dominant service providers.
• Enable European broadcasters and TV operators to establish themselves as recognized players in the Targeted Advertising ecosystem by enabling contextual advertising and user profile enrichment.
• Allow broadcasters and operators with limited innovation budgets to significantly improve their services.