Title: Contextual Bandits for web page conversion rate optimization
Abstract: Unbounce is one of the world's leaders in providing tools to create landing pages and drive conversions for marketing teams and agencies. We are constantly striving to use machine learning to deliver more value to our customers by increasing the number of clicks, conversions and leads they receive. Contextual bandit algorithms are an online machine learning approach in reinforcement learning that can be used to dynamically allocate traffic to several variants of a webpage to optimize a metric, such as conversion rate on the page. This allocation depends on both the page variants as well as the properties of each visitor, allowing visitors to be directed to the page variant most likely to perform well for that visitor in real time and delivering significantly improved conversion rates for pages.
Over the last year we have built a proof of concept contextual bandit system for visitors to Unbounce hosted landing pages. In this talk we will describe how we developed bandit routing algorithms and evaluated the system to estimate the impact it would have on page conversion rates.
Bios:
Thomas Levi ( https://twitter.com/tslevi , https://www.linkedin.com/in/thomas-levi-40a09432/ ) started out with a doctorate in Theoretical Physics and String Theory from the University of Pennsylvania in 2006. His post-doctoral studies in cosmology and string theory, where he wrote 19 papers garnering 900+ citations, then took him to NYU and finally UBC. In 2012, he decided to move into industry, and took on the role of Senior Data Scientist at POF. In 2015, he became Director of Data Science at Unbounce. Thomas has been involved in diverse projects such as behaviour analysis, social network analysis, scam detection, Bot detection, matching algorithms, topic modelling and semantic analysis.
Jordan Dawe ( https://twitter.com/freedryk , https://www.linkedin.com/in/jordandawe/ ) is a data scientist and Python developer with a PhD in computational climate physics from the University of Washington. After working as a Postdoc at UBC he was hired by the website DeviantArt in 2013, where he worked on search rank metrics, traffic log analysis, topic models, and classifier systems. In 2016 he became a Senior Data Science Developer at Unbounce, working on the research and development team.
Michael McDermott ( https://www.linkedin.com/in/michael-mcdermott-413a24a7/ ) received his PhD in Theoretical High Energy Physics from UBC in 2014 during which he made contributions to our understanding of entanglement. He then co-founded a data science consulting company called MLinfinity which helped many local startups with complex needs. He has since been working as a data scientist exclusively in tech, joining POF in 2015 and most recently Unbounce in 2017. He has worked on a wide array of problems including deep convolutional learning, recommender systems, anomaly detection, bayesian hypothesis testing and the mathematics of virality.
Schedule:
- 6:00PM Doors are open, feel free to mingle
- 6:30 Presentation start
- ~7:45 Off to a nearby restaurant for food, drinks, and breakout discussions
Getting There:
By transit there a number of high frequency buses (check Google Maps or the Translink site for your particular case) that will get you there. There is also a SkyTrain station within walking distance.