Using Causal ML to select articles for promotion on the homepage and optimise article performance
Join the Data Science Expert GroupNZZ, Süddeutsche Zeitung: Using Causal ML to select articles for promotion, optimise content and paywall performance.
A common decision-making problem for digital content publishers is deciding which content to promote on their distribution channels (e.g., homepage, email newsletters, social media), where personalisation is often undesirable or impossible. NZZ and ETH developed a scalable and robust learning prototype framework using Causal ML to support this decision-making and curation challenge using real-world data from NZZ and a Python-based causal ML package from Microsoft Research. Dr. Cristina Kadar, Data Science & Machine Learning Product Owner at NZZ, and Joel Persson, Research Associate and PhD candidate at the Chair of Technology Marketing at ETH Zurich, will discuss how the framework outperforms traditional curation methods and significantly improves reader engagement. The research paper behind this experiment at NZZ is available online.
For its part, Süddeutsche Zeitung leverages causal ML methods to increase its subscription base. One crucial question for conversions is how many articles a customer should be able to see before hitting the paywall (“metering limit”). Felix Hagemeister, Data Scientist at SZ will explain how they use Microsoft’s econML Python library to fit a causal tree to decide personalised metering limits automatically. The causal tree optimises the conversion probability and splits customers into groups with higher or lower metering limits depending on user behaviour. Other applications of causal inference methods they harness are high-dimensional fixed effect models. If AB-testing is not feasible, these models can better guide decision-making problems (such as which feature to extend or which service to prioritise) than simple correlations.
WHEN: November 21, 2023, from 2 pm to 3.30 pm CET
WHERE: REGISTER HERE for this free roundtable discussion powered by the WAN-IFRA Data Science Expert Group.