The research chair KWANKO-EMLV-ESILV aims to contribute to the reinforcement of the predictive power of purchasing models or advertisement impact by analyzing data from social networks and online communities.
French company Kwanko is the European leader in multi-channel advertising, with an extensive background in Online Performance Advertising. David Dupuis, an ESILV graduate, is the first Data Scientist researcher to participate in the KWANKO-EMLV-ESILV chair. A few months after the start of his collaboration with Kwanko, he gave us insight on his PhD research progress.
Joining the chair
I joined the KWANKO-EMLV-ESILV research partnership in September 2016 and officially started my research with the CNAM-EDITE in February 2017.
My preliminary work focused on learning as much as possible on the topic of real-time bidding (RTB), a modern way of adaptive advertisement placing in real-time. I was also helping Kwanko set up their Data Management Platform (DMP) which would be the foundation of my work and further Research & Development within the company. The DMP is an advanced tool for intelligently gathering, storing and analyzing big and complex data generated by online advertising in order to extract meaningful information and turn it into actionable insights.
Gradually, I turned toward the analysis of influence maximization in the context of RTB.
What is real-time bidding?
RTB consists in targeting users individually with appropriate advertisements when they open a web page. The process involves tracking the users with cookies and Supply-Side Platforms selling the available ad slots to ad agencies (Demand-Side Platforms) in a real-time bidding process that follows what is known as a Vickrey auction.
Since each user belongs to a community whom he or she interacts with through word-of-mouth or through online social networks such as Facebook or Twitter, we aren’t just targeting a user but his or her friends as well. Thus, each user is a potential influencer in his or her network.
The idea of influence maximization in a viral marketing strategy is to tap into this power of influence to maximize the spread of individuals being affected by the information.
Doing this with ads means increasing the potential effect of a single ad targeted to a single user. Business wise, you could target more people while spending less money.
Influence maximization in the context of real-time bidding
The topic of influence maximization has been widely studied over the past decade with some well-known names such as Domingo & Richardson(1), David Kempe, Jon Kleinberg(2), Wei Chen(3), Yuchen Li(4), Francisco Bonchi(5), Laks Lakshmanan(6) and many more. These groups of researchers have elaborated novel ways to effectively find within a social network seed sets of individuals to maximize influence spread. However, to our knowledge little to no research has been done taking into consideration technical constraints of real-time bidding for display advertising. This means making smart targeting decisions in less than 100 milliseconds and without the freedom of targeting any user. Or, the obvious concern for data privacy that makes third party advertising agencies unable to directly obtain social network information on the user. It is also challenging to analyze the influence of an ad on a user and even more so the influence of this user on his peers.
Undoubtedly, research on influence maximization in real-time bidding is at a very theoretical stage but this is also the beauty of scientific research: “to boldly go where no man has gone before”.
(1) Mining the Network Value of Customers – 2001 – P. Domingos, M. Richardson
(2) Maximizing the Spread of Influence through a Social Network – 2003 – D. Kempe, J. Kleinberg, E. Tardos
(3) Scalable Influence Maximization for Prevalent Viral Marketing in Large-Scale Social Networks – 2010 – W. Chen, C. Wang, Y. Wang
(4) Real-Time Influence Maximization on Dynamic Social Streams – 2017 – Y.Wang, Q. Fan, Y.Li, K-L. Tan
(5) Learning Influence Probabilities in Social Networks – 2010 – A. Goyal, F. Bonchi, L. V. S. Lakshmanan
(6) CELF++: Optimizing the Greedy Algorithm for Influence Maximization in Social Networks – 2011 – A. Goyal, W. Lu, L. V. S. Lakshmanan