wrap up ebu big data and society conference at rtbf - day 2 (13 december 2016)
TRANSCRIPT
BIG DATA & SOCIETY WORKSHOPDAY 2 INTRODUCTORY SPEECH
Pierre-Nicolas Schwab, Big Data/CRM Manager
13 December 2016
RTBF
• Personalization is key: PSM can’t do without
• We have values that need to be reflected in our algorithms
• Sharing knowledge among EBU members is key for advancement
MAIN IDEAS COVEREDYESTERDAY
• Ecosystems and modeling
• Big Data in the newsroom
• A model to understand audience engagement
• History of recommendation and filter bubbles
• Ethical recommendations
MAIN IDEAS COVEREDYESTERDAY
BigData & Society
Serendipity
Customer Value
Ethics
Public Service Medias
Filterbubbles
Person-nalization
Infobesity
• Modeling comes first (remember the many stakeholders in the electrical case ?)
• You can’t really understand the world of you don’t model (in opposition with most Big Data practices today)
• What is your strategy? (Short-term vs. long-term goals)
TALK 1 : PROF. WEHENKEL
• What is PSM’s position (data-centric firms vs. governments vs. all other industries) what is our vision for Big Data ?
• Are our projects aligned with trends in Science / Tech / Producers / Consumers
• Relevancy / velocity of data sources
TALK 1 : PROF. WEHENKEL
• Content creation can be supported by Big Data technologies
• Big Data can be used at each step of the flow :– Discover
– Create
– Curate
– Engage
TALK 2. STEVEN BOURKE(SCHIBSTED)
• Meta data is essential (holy grail !)
• Content creators can contribute better meta data
• How do we create value for content creators (metrics, ease-of-use)
• Balance human / algorithmic curation
TALK 2. STEVEN BOURKE(SCHIBSTED)
• Engagement is what we strive for : but do we know what it is really ?
• Engagement model proposed (scientifically validated) :– Brand perceptions
– Brand dialog behaviors
– « Shopping » behaviors
– Brand consumption (RFV)
TALK 3. PROF. MALTHOUSE(NORTWESTERN UNIVERSITY)
• Output of model : SAT, LOY, Lifetime value, customer value (remember CLV?)
• Key takeaways :
– where do we get data for those 4 components to measure total value created
– How many « ecosystems » do we have?
TALK 3. PROF. MALTHOUSE(NORTWESTERN UNIVERSITY)
• Recommendation algorithms are not new (1992)
• They have fundamentally changed in nature and have become more complex
• Data collection has changed:
– Explicit Implicit
– Non intrusive intrusive
TALK 4. PIERRE-NICOLAS SCHWAB (RTBF)
• Two kinds of traps :
– Content trap (filter bubble)
– Ecosystem trap
• How do we create value for user: E³
– Educate
– Encourage
– Excite
TALK 4. PIERRE-NICOLAS SCHWAB (RTBF)
• Recommendation algorithms can go very wrong (racist, sexist, discriminatory)
• Algorithms reflect beliefs of our society
• When we design recommendation systems we must identify those who may be negatively impacted
TALK 5. EVAN ESTOLA(MEETUP, NEW-YORK)
• Gender may become a discriminatory factor : identify it and remove it from your model
• Differentiate good / bad / horrible feature
• use ensemble modeling
TALK 5. EVAN ESTOLA(MEETUP, NEW-YORK)
• All presentations already made available for your comfort on slideshare : www.slideshare.net/intotheminds
• I’ll redo the presentation with Evan and make it available as video file for your comfort
ONE LAST WORD