Online Program

Dynamic Clustering of Subscribers at the Los Angeles Times (308538)

Keywords: Dynamic clustering, network-based clustering, subscriber data

In this talk I will discuss how our data science team has used dynamic clustering techniques to understand our subscriber base and find drivers of retention. I will focus on two related projects, and the common challenge of modeling unstable populations:

1) Dynamic clustering of subscribers. We used a hidden Markov model with time-dependent parameters to cluster subscribers over time and allow for group-switching; 2) Classification of subscribers’ topical interests over time using network-based methods. Due to the dynamic nature of article data, we use a network derived from user similarity for robust temporal classification. I will discuss the special challenge of understanding user interests when most news is focused on one topic: coronavirus.

These projects have fed into an overarching goal of growing our subscriber base.