An Interview with the Winners of IBC’s Best Conference Paper Award 201705 Sep 2017
This year the IBC Best Conference Paper has been awarded to Marcelo Souza and his colleagues Joāo Castellani, Daniel Monteiro and Carlos Octávio Queiroz from Brazilian broadcaster TV Globo, for their paper entitled ‘Big data for data journalism, enhanced business analytics and video recommendation at Globo’.
IBC interviews the winning authors.
Tell us a little about the team. What special skills did you each bring to the project?
Marcelo Souza is Head of Digital Media for the Technology Division at TV Globo, responsible for delivering initiatives related to OTT, Video Syndication, Web Tools for TV Production, Business Intelligence and recently adTech. After 16 years at Globo, Marcelo has developed skills across diverse areas including telecommunications project management, post production, dth and cable delivery for the international channel, budget planning and technology business partnership.
João Castellani leads Big Data initiatives at TV Globo, and has been working on Architecture for Information Systems for the past 20 years. He has expertise on the complete big data and analytics value chain, from foundation, data ingestion, data preparation, data intelligence to data delivery, based on cloud environments (Amazon AWS, Google Cloud, Microsoft Azure, and so on) and tools such as Haddop, Hive, Spark and Apache NiFi.
Daniel Monteiro leads TV Globo’s R&D division and has been working on Innovation Management for the past 17 years. He has expertise on TV production and distribution workflows, Software Engineering and Agile Development, Recursive Algorithms and Neural Networks. His latest project is called “Powered by Respect”, where a tetraplegic person was able to drive a formula 1 car using brain wave commands through a sensorized helmet and an algorithm behind, won a Silver Lion on the Lions Innovations / Creative Data at Cannes Lions – Festival of Creativity.
Carlos Octávio is Head of Business Partnership and Architecture for the Technology Division at TV Globo with 19 years of experience in the Media and Entertainment industry. He has solid skills in Information Technology, acquired in large size companies in executive positions, working in the following areas: IT Strategy and Governance, Technology, Information Security, System Development and Integration, IT Operations and Infrastructure.
Is this the first project on which you have worked together?
Yes and no. Although we have worked in some isolated projects before, this is the first project where we brought all skills together, from Engineering and IT groups, that now are under the same umbrella of Technology at Globo. This multi-disciplinary set of skills was fundamental for the project success.
The application of big data analytics across your company is ambitious and exciting from both a technological and organisational viewpoint. When did you start the work and how did you decide the rate at which you would roll it out into the different departments?
We have been studying big data and machine learning for more than a decade, including neural networks, distributed computing and file systems, and so on, however it was only last year that TV Globo structured an exploratory team to build its first TV business applications using big data and machine learning tools. It’s important to say that Globo.com, our digital company inside Grupo Globo, has been working with these tools for about 4 years, but it was focused only on our websites.
How did you manage the introduction of radically new ways of working and staffing? Was there any resistance to the changes?
Our strategy was to classify our efforts as exploratory. By doing this, our internal clients saw this as an innovation driver, rather than an operational one. Also, for this first phase, we have elected areas that were more mature with data understanding such as Journalism, Intelligence and Digital Media, so resistance was minimum. Actually, we had to hold on expectations at first, which is a great problem to deal with as an engineer. After the great results, it was easier to move to a more operational mode, if such thing exists in the “always beta” world we live in.
How have the overall running costs of the analytics-based departments changed since you introduced data-analytical working?
This is yet to be seen in the long run, however the cost of data extraction has changed a lot since, for traditional Business Intelligence projects, we have to pay extra effort for every new cube on the data warehouse, which for our big data project there is one cost to bring data to the data lake and all analysis is run on top of it. Also, we have internalised human resources to make developments more agile, so the overhead of third party was gone.
What other departments in your company do you think would benefit from a data-analytical approach?
All of them, from Management to Content Production. Sports are a natural candidate, since sports data is largely available. The most challenging area is Entertainment, where subjectivity and artistic evaluation have the most value. We would like to amplify Globo Studio’s to a more data driven TV production, with big data and machine learning applications that would go deeper on analytics for narratives, authors, characters and casting.
You highlight many positive factors in the move towards a data-aware organisation but have you discovered any pitfalls which others should avoid?
Data quality and data governance are two dangerous topics that could ruin any data driven organisation, for me, those are the pitfalls that we have overcome in the beginning. You have to be very strict with homologation process from data lake ingestion to the data exposure for applications, and have one specialised team that the company trust to lead the process.
What future advances would allow you to improve the performance of your data analytics systems?
The continuous migration of big data foundation to open source frameworks and cloud environment is very important to guarantee transparency, agility and performance. Self-data preparation is something the we are looking to implement in the near future. I would say that data lake metadata management is a field yet to be explored by the industry.