ECML/PKDD’22
Uplift Modeling Tutorial & Workshop

(c) Laurent Espitallier CC by-sa 3.0

Topic

Uplift modeling concerns the data-driven estimation of individual treatment effects for optimizing or customizing decision-making in business. Uplift modeling is receiving a growing interest, both in academia and industry with growing number of applications in marketing, pricing, learning analytics, operations management, etc. Uplift modeling draws from both the field of causal inference and machine learning, is closely related to causal effect estimation (conditional average treatment effect, heterogeneous treatment effect, individual treatment effect) but is strongly application oriented. For example, unlike most research on causal discovery, uplift modeling puts emphasis on randomized trials which are commonly available in the industry (e.g. A/B testing), ranking based performance measures and taking into account costs. A large number of open research questions are still to be addressed and the domain would benefit from both formalization as well as closer integration with the field of causal effect estimation.

The first Uplift Modeling Tutorial and Workshop at ECML/PKDD'22 aims at gathering the uplift modeling research community and will bring together experts from both academia and industry. The tutorial aims at providing a broad but concise overview of the state-of-the-art in uplift modeling and highlight challenges and directions for future research. The workshop provides the opportunity for researchers and developers to present new approaches for uplift modeling, business applications and data sets, and to discuss on open issues and connections with related research fields. An invited talk by Eustache Diemert, Senior Staff Research Lead at Criteo AI Lab at Grenoble, on Uplift Modeling for Online Advertising will kick-off the workshop and provide an industry perspective towards uplift modeling.

Keynote Speaker

Eustache Diemert, Criteo AI Lab, France
Title: “Uplift Modeling for Online Advertising”

Eustache Diemert is a Senior Staff Research Lead at Criteo AI Lab at Grenoble. He works on various aspects of machine learning such as causality, uplift modeling and privacy.

Call for papers

The uplift modeling tutorial and workshop at ECML/PKDD2022 aims at bringing together researchers and practitioners working on uplift modeling, to present and discuss on recent developments, to identify open issues, and to form a community and foster future initiatives. We welcome submissions that present original contributions related to uplift modeling, or, more generally, to the use of causal effect modeling (e.g., conditional average treatment effect, individual treatment effect, heterogeneous treatment effect) for decision support.

Topics include but are not limited to:
  • Uplift modeling methods
  • Evaluation of uplift models
  • Practical aspects of causal discovery and effect modeling
  • Causal discovery and effect modeling under biased treatment assignment
  • Beyond binary treatments, e.g., continuous, multi- and high-dimensional treatments
  • Beyond binary outcomes, e.g., continuous, multiple and time-dependent outcomes
  • Including costs and benefits in the uplift modeling process
  • Learning to rank for uplift modeling and causal effect estimation
  • Business applications of uplift modeling and causal discovery and effect estimation
  • Descriptions of datasets

Submission

Submitions should be made HERE.

Various types of submissions are encouraged: full papers and extended abstracts. We also make it possible to directly submit a poster you intend to present at the workshop.

Papers and abstracts should adhere to the general paper formatting guidelines as provided on the ECML/PKDD2022 conference website.

Authors of accepted submissions will be invited to present their paper in an oral presentation (accompanied by a poster) or a poster presentation.


Please note that at least one author of each accepted paper should register for the conference. Videoconference registration is acceptable.

Key dates

Workshop paper submission deadline: 20 June 2022 30 June 2022
Workshop paper acceptance notification: 13 July 2022 15 July 2022

Program Committee

Chairs

  • Szymon Jaroszewicz, Polish Academy of Sciences & Warsaw University of Technology, Warsaw, Poland
  • Wouter Verbeke, KU Leuven, Belgium

Members

  • Mouloud Belbahri, TD Bank Group, Canada
  • Artem Betlei, Criteo, France
  • Jeroen Berrevoets, University of Cambridge, United Kingdom
  • Kristof Coussement, IESEG School of Management, Lille, France
  • Eustache Diemert, Criteo AI Lab, France
  • Robin Gubela, Humboldt University of Berlin, Germany
  • Leo Guelman, Royal Bank of Canada, Canada
  • Stefan Lessman, Humboldt University of Berlin, Germany
  • Diego Olaya, ACA Group, Belgium
  • Krzysztof Rudaś, Polish Academy of Sciences, Poland
  • Piotr Rzepakowski, Warsaw University of Technology, Warsaw, Poland
  • Sam Verboven, Vrije Universiteit Brussel, Belgium

Tutorial

Outline

The tutorial aims to provide a broad overview of the state-of-the-art in uplift modeling. Building on a concise introduction to establish a clear problem definition and to identify application domains, the tutorial will discuss the most prominent modeling methods, including decision trees, ensembles, meta-learners and deep learning. Moreover, the evaluation of uplift models will be thoroughly discussed, since it poses a particular challenge resulting in a variety of approaches being used in practice. Finally, the tutorial also aims to provide a view on available implementations (open source packages), on recent developments in learning from observational data, and an overview of recent literature and open issues (which will serve as input to the panel discussion in the workshop).

Instructors

Szymon Jaroszewicz

Prof. dr. Szymon Jaroszewicz is a full professor at the Institute of Computer Science of the Polish Academy of Sciences, and at the Faculty of Mathematics and Information Science of Warsaw University of Technology, Warsaw, Poland. His main research interest is causal discovery and in particular uplift modeling. He is an author of several papers in this area. Several of the uplift modeling algorithms he co-developed are widely available in popular uplift modeling packages. For several years he has been a member of programme committees of major AI and Machine Learning conferences such as SIGKDD, IJCAI, ECML/PKDD.

Contact: s.jaroszewicz@ipipan.waw.pl

Website

Wouter Verbeke

Prof. dr. Wouter Verbeke, is an associate professor of data science at the Decision Sciences and Information Management Department of KU Leuven in Leuven, Belgium. His research is situated in the field of cost-sensitive and causal machine learning for business decision-making, with a strong focus on the development of uplift modeling and applications in operations management, pricing, credit and fraud risk management, and marketing. He is member of the scientific program committee of IFORS, the three-yearly global conference on operations research, and yearly organizes sessions and streams at the EURO conference by the European Operations Research Society.

Contact: wouter.verbeke@kuleuven.be

Website

Workshop

The workshop aims to provide a forum for the research community to present and discuss the latest developments in the field. The workshop will take off with a keynote by Eustache Diemert, Senior Staff Research Lead at Criteo AI Lab at Grenoble, an expert in the field of uplift modeling who is working in the industry and will provide a practitioner's view on uplift modeling, with a particular focus on challenges and recent developments. After the keynote, a session with presentations or papers are foreseen, with a coffee break in between. The workshop is to trigger discussion on open issues in uplift modeling and directions for future research, and aims for active involvement of the workshop participants. Since uplift modeling is an emerging area of research, a key objective of the tutorial and workshop is to bring together (for the very first time) and to establish an uplift modeling research community, to launch initiatives for future gatherings (this workshop is envisioned to be a starting point by the chairs) and to foster collaboration among researchers in the field.

Schedule (tentative)

The tutorial and workshop will take place on Monday September 19th in the afternoon.
Tentative schedule and outline of the tutorial and workshop:

  Tutorial
  30 min. Session 1 by Szymon Jaroszewicz
  • Introduction to uplift modeling
  • Uplift modeling methods 1 (decision trees, ensemble methods)
  • Evaluation of uplift models 1
  • Implementing uplift models: packages
  •   30 min. Session 2 by Wouter Verbeke
  • Uplift modeling 2 (meta-learners, deep learning, learning2rank)
  • Evaluation of uplift models 2
  • Learning from experimental and from observational data
  • Review of the literature and open issues
  •   Workshop
      45 min.   Invited talk by Eustache Diemert, Criteo AI Lab, France: Uplift Modeling for Online Advertising
      30 min.   Coffee break
      60 min.   Paper Presentations