Experimentation to Productization : developing a Dynamic Bidding system for a location aware Mobile landscape
This session is to help structure a Hypothesis based approach to Engineering problems and learning to quickly translate & implement algorithms on weblogs(mobile footprints) data.
This session is about 2 main things –
1. Introduction to a Real Time Bidder(RTB) & Dynamic bidding in a location based mobile marketing
2. Three specific problems that we addressed to increase the bottom-line for our clients & how we scaled them.
I will start with the obvious, ie. how do advertisers reach you on mobile, fetching your comprehensive digital footprint, all in 6 milliseconds, or less. Then look through sample digital footprint(weblogs), laying the ground to understand the data, and algorithms to derive statistical relationships.
Towards this, I will then talk about identifying quick wins to deliver outcomes, all through data and in this introduce Hypothesis based Engineering - ie how not to go down a bottomless pit.
I will then spend majority of time talking about 3 problems we adressed at AdNear to increase the bottom line for our clients -
Algorithm to develop a Dynamic bidding system which prices each opportunity to bid, based on the quality of that “specific” inventory - Towards this, I will focus on how we built meta-data for otherwise, not so userful attributes like “user-agent”, data for creatives, besides the obvious “features”
Characterizing user-mobilty patterns to generate user profiles - ie given a cross section of user, how do we map the activities associated with their geographical footprint - and generate & probablistic picture of his activity patterns & affinity towards general activities
Developing a comprehensive app-ranking system : How we use web to increase the information content of the apps to deliver Business outcomes that matter. The system updates the snapshot across multiple dimensions for each of the unique appids in the system every hour, to deliver a self aligning machine learning system at scale
Finally I will close this with the framework we built to measure all this in real time - A/B testing framework, Simulation & Reporting - which supported the Experimentation phase, created stickness that pushed the productization of data into our Production systems, while doing so at Scale.
General familarity with Real time bidding(RTB), Mobile targeting, Information retrieval & ranking systems, is preferble, though the talk will be armed with all that is needed to get though.
Ekta is Data Scientist with AdNear Pte., where she is designing Dynamic Bidding systems and A/B testing framework for bidding in location based mobile targeting space, to increase the bottom line for clients across Asia-Pacific. She has a background in Quantitative Economics(MS) from Goethe-University, Frankfurt and Computer Science(BS) from Bangalore, India and enjoys Monetizing and leveraging technology to solve abstract Business problems. While at Grad school she became passionately interested in rationality, framing problems and how we human being respond to ambiguous choices, something she sews in technical dimensions with a scientific rigour.
Prior to AdNear, she was with 7 Inc., Innovation Labs, where she was responsible for end to end solutioning, statistical analysis and deployment of Analytic models for e-commerce clients and designing intuitive customer experiences. Before that she has worked in roles across Quality Engineering (VMware Inc.), Program Management (SAP Labs) and Experimentation methods, Auctions & Macroeconomics while pursuing her Masters at Goethe University.
She presented a talk at Pycon 2013, Bangalore, selected as a speaker for Pycon APAC, 2014 Taipei. Also accepted to present the same in Grace Hopper’s conference for Women in Computing, 2014 at Pheonix(USA)
- Linkedin - http://www.linkedin.com/profile/view?id=38210724
- Github - https://github.com/ekta1007
- Links to Pycon 2013 talk -
- Slides @ slideshare : http://www.slideshare.net/ekta1007/pycon-2013-experiments-in-data-mining-entity-disambiguation-and-how-to-think-datastructures-for-designing-beautiful-algorithms