US 12,137,148 B1
System and method for estimating a long-term effect in the presence of unobserved confounding
Ciarán Gilligan-Lee, Dublin (IE); Lucas Maystre, London (GB); and Graham Van Goffrier, Leicester (GB)
Assigned to Spotify AB, Stockholm (SE)
Filed by Spotify AB, Stockholm (SE)
Filed on Jul. 13, 2023, as Appl. No. 18/351,893.
Int. Cl. H04L 67/306 (2022.01); H04L 67/00 (2022.01)
CPC H04L 67/306 (2013.01) [H04L 67/34 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for estimating a long-term effect of a change in user interface of a media playback application, the method comprising:
compiling, at a server, an experimental dataset including short-term data on the change in user interface using data from a first plurality of computing devices executing the media playback application, wherein the short-term data includes a treatment and a short-term outcome and is associated with a first time period;
determining from the experimental dataset, a short-term effect of the treatment on the short-term outcome;
compiling, at the server, an observational dataset including the short-term data on the change in user interface and long-term data on the change in user interface, wherein the long-term data includes a long-term outcome and is associated with a second time period that is greater than the first time period, and wherein the observational dataset includes unobserved confounding, the server compiling the observational dataset using data from a second plurality of computing devices executing the media playback application; and
estimating a long-term effect of the short-term outcome on the long-term outcome, wherein the long-term effect of the short-term outcome on the long-term outcome is estimated by:
computing a portion of the short-term outcome using samples from the observational dataset and the short-term effect, wherein the portion of the short-term outcome has no causal contribution from the treatment, and
performing regression using the portion of the short-term outcome.