#	Pagina
attuale pagina	/open-h2020/projects/213154/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - ReTV (Enhancing and Re-Purposing TV Content for Trans-Vector Engagement)

Teaser

Summary

ReTV\'s objective is to re-invent TV for the interactive age. We aim to provide broadcasters and content distributors with technologies and insights to leverage the converging digital media landscape. Viewing of linear broadcast TV is decreasing while time spent with digital content on Catchup TV, on-demand OTT or social media rises. Broadcaster audiences are fragmented across digital channels, which are full of competing content. The TV industry has to catch up with online competition in web technology: user tracking, personalisation and targeting. ReTV will provide tools to compete in the digital ecosystem and maximise the success of content publication. The Trans-Vector Platform (TVP) will provide stakeholders with the ability to â€œpublish to all media vectors with the effort of oneâ€. The Trans Vector Platform is a collection of systems that covers all use cases of Trans Vector Publishing. By using a modular architecture the TVP can easily be extended, integrated into existing workflows and create synergies between systems.
- The TVP understands content - it has knowledge of genres and program topics and the concepts inside programs
- The TVP understands propagation strategies - it measures, visualizes, and reports on content engagement and viewer patterns across all tracked vectors, and determines a propagation strategy based on prediction models and machine learning.
- The TVP knows how to recommend content to viewer cohorts (i.e. groups of viewers of a similar age, gender, location, history, or device/vector) - items are automatically matched and timed with the audience on the publication vectors or recommended via TVP-powered services on the user-facing application
- The TVP guides content re-purposing decisions - it can generate automated suggestions for re-purposing of content assets for more reach and engagement on the published vectors, e.g. video summarization for social media distribution.

Work performed

In M1-18 we established the initial Trans Vector Platform (TVP) prototype, integrating components and services for data collection, annotation, analysis, query and visualisation. This platform enables data-driven solutions to provide content re-purposing and recommendation functionalities in order to optimise the publication of media content on the Web and social media. Four scenarios have been implemented as proof of concepts and tested with professional users (media organisations such as TV broadcaster) and consumers (TV viewer).

WP1: the data collection and annotation pipeline of ReTV. We have collected to date almost 20 million documents of various types (web pages, social media documents, EPG entries). For video annotation, we developed and set up a service that performs temporal fragmentation and fragment annotation using up to 6 concept pools totaling over 6000 concepts. For brand detection, we developed and set up a service that detects, to date, more than 300 logos/brands; and, combining the annotation and brand detection results we also developed an ad detection method. For ensuring all this metadata is recorded, stored and communicated in a way that facilitates their subsequent use in the ReTV applications, we specified and implemented the necessary formats. To extract additional metadata from the text in support of data analytics, we set up and configured NLP and NER processes for text analysis.

WP2: a first version of the ReTV predictive analytics (due M20). The extraction and provision of past data regarding events, content-based success as well as TV audience is necessary to train the prediction models. The first, individual prediction experiments with the available data sets have set a baseline for the foreseen hybrid prediction solution as well as have provided insights to the use case partners and an opportunity to evaluate professional userâ€™s expectations for prediction in their user tests.

WP3: a complete pipeline that allows the use cases to support content adaptation and re-purposing as well as content recommendation and scheduling.

WP4: the planning of the system architecture and the seamless integration of all technical components into a first complete prototype of the TVP. To ensure scalability we follow a flexible, distributed content processing strategy. The TVP Visual Dashboard has been deployed, synchronizing multiple data visualization methods.

WP5: successfully and timely performed user tests with media professionals that supported the design of the content owner use cases and the implementation of the TVP. To gather sufficient qualitative data, task partners performed in-depth interviews with selected professionals from the target groups. For quantitative data, surveys and questionnaires were distributed to a wider range of professionals from various European organisations.

WP6: the definition and generation of the consumer Use Case scenarios. Requirements were created with the help of questionnaires given to people during IFA 2018 and and also distributed online. Based on this feedback, we created the first mock-ups and wireframes for three scenarios. Clickable wireframes were tested with a small group of consumers. With this feedback, the first prototypes were built.

WP7: project partners successfully promoted the project and grew their audience via digital comms channels, exceeding the target impact numbers in cases. The ReTV Stakeholder Forum and Advisory Board were established with respectively 27 and 5 members. The partners presented the project at five events (covering all of the projectâ€™s target groups) and co-organised a workshop that specifically targets research questions and audiences at the core of the project (DataTV at ACM TVX 2019).

WP8: effective project, administrative and financial management of the project from day one to the time of writing.

Final results

In ReTV, we are building a unique solution for broadcasters: the TVP. To achieve this, we go beyond the state of the art with respect to: video analysis and annotation, data annotation (Named Entity Recognition and Linking), data analytics for TV content, predictive analytics for TV audiences and content success, tools for professional users to select, adapt and optimally publish content, services for TV consumers with personalised content recommendations. We are going to build on the GENUINE interest in the TVP and turn it into uptake, with these expected results until the end of the project:
- Fine tuning our predictive analytics, video summarization and content recommendation algorithms
- TVP Visual Dashboard update with advanced data visualisations (audience, WYSDOM)
- Topics Compass and Content Wizard extended with prediction capabilities, longer testing with professional users
- Content sWitch and 4u2 Chatbot demoed publicly and tested extensively
- Increased scientific output and public dissemination activities
- Engagement with stakeholders with the aim of gaining early adopters
- Business planning with an eye to post-project exploitation
The impact will be to future-proof European broadcasters, who need to adapt to the new digital content ecosystem and compete with new pure-play and OTT competitors. Right now, they lack the right tools for this - the tools that ReTV is going to provide through the Trans-Vector Platform.