Mastering Uplift Modeling for Better Recommendations

## RecSys IL Meetup #3

Mastering Uplift Modeling for Better Recommendations

Data science often hinges on the analysis of correlation, but the power to understand and predict outcomes significantly increases when we turn our gaze towards causality. In this meetup, we delve into the complex, yet compelling world of Causality and Uplift Modeling, and their profound impact on Recommendation Systems.

We are grateful to eBay for their community sponsorship and to Gong for hosting this event.
Talks are in English.

Agenda:
18:00 - 18:30 Registration, Mingling, Pizza & Beer
18:30 - 18:45 Opening, Host and Organizer words.
18:45 - 19:15 "How to make better decisions - causality, uplift and everything in between״
Ohad Levinkron-Fisch, VP AI at DealTale (Vianai company)
19:15-19:45 "Leveraging Causal Uplift Modeling for Budget Constrained Benefits Allocation"
Dima Goldenberg, Senior ML manager at Booking.com
19:45-20:15 "Uplift models evaluation"
Michael Kolomenkin, Head of AI Research at Playtika

Abstracts:
1. How to make better decisions - causality, uplift and everything in between
Much of our work as data scientists is about using data to make decisions: who should I send this offer to? Should we insure this client? Which movie should we recommend? In this session we'll review why causality is needed for decision making, and drawbacks of other common approaches (e.g. "predict then optimize"). We'll walk through how to build decision making policies and validate them. The focus will be on cases when there's enough experimental data to train as well as validate ("uplift modeling"), but we'll also mention observational studies and how they improve business KPIs in real life. Finally, we'll understand how Reinforcement Learning is related to causality and where the state of the art is at.

2. Leveraging Causal Uplift Modeling for Budget Constrained Benefits Allocation
In the e-commerce industry, promotional offers such as discounts and rewards became a key driving force to encourage customers to complete purchases. However, these offers may also lead to monetary losses for businesses. Causal machine learning and optimization methods can be utilized to drive the allocation of promotions, while efficiently controlling spending within a given budget.
In this case study, we will share the promotion assignment techniques that our team at Booking.com has developed and implemented in real business setup. We will describe dedicated methods for uplift modeling under budget constrains while learning from limited data. By leveraging these techniques, we were able to estimate the effect of discount offers and allocate them in a real-time budget constrained setup, enabling us to unlock promotional campaigns while bringing more value to our customers and simultaneously growing our business.

3. Uplift models evaluation
While uplift models may utilize traditional regression and classification techniques, they present distinctive hurdles and are not suitable for evaluation with standard supervised-learning methodologies. In this presentation, we will elaborate on the intricacies of the uplift modeling issue, showing how to assess and characterize the effectiveness of these models.
The evaluation techniques discussed during this session have been integrated by Playtika into an open-source Python package.

Speakers:
Ohad Levinkron-Fisch is VP AI at Dealtale (Vianai), building the world's first platform for provably improving business decisions with causality. Previously, Ohad served as Head of Data Science at Clalit Research, which he transformed into a world leading center for causal inference in medicine, collaborating with the top researchers in the field. He holds an MSc in Physics from the Weizmann Institute, and BScs in Physics and EE from Ben Gurion University.

Dima Goldenberg is a Senior Machine Learning Manager at Booking.com, Tel Aviv, where he leads ML efforts in promotions personalization, pricing, applied research and machine learning excellence.

Michael Kolomenkin: A scientist, inventor, and technology visionary, Michael is passionate about making life better with technology and particularly AI. He has almost twenty years of experience in developing and leading successful projects in diverse technological areas. He co-founded several companies in the past and holds a Ph.D. in EE from the Technion. Currently, Michael is leading the AI research activities in Playtika