US 11,790,379 B2
Bayesian estimation of the effect of aggregate advertising on web metrics
Shiv Kumar Saini, Karnataka (IN); Ritwik Sinha, Cupertino, CA (US); Moumita Sinha, Cupertino, CA (US); and David Arbour, San Jose, CA (US)
Assigned to ADOBE, INC., San Jose, CA (US)
Filed by ADOBE INC., San Jose, CA (US)
Filed on Aug. 27, 2020, as Appl. No. 17/4,377.
Prior Publication US 2022/0067753 A1, Mar. 3, 2022
Int. Cl. G06Q 30/0201 (2023.01); G06Q 30/0204 (2023.01); H04N 21/81 (2011.01); G06Q 30/0242 (2023.01); H04L 67/50 (2022.01); G06N 7/01 (2023.01)
CPC G06Q 30/0201 (2013.01) [G06N 7/01 (2023.01); G06Q 30/0205 (2013.01); G06Q 30/0242 (2013.01); H04L 67/535 (2022.05); H04N 21/812 (2013.01)] 13 Claims
OG exemplary drawing
 
1. A computer-implemented method for data analytics, comprising:
training, by a processor, an artificial neural network to model a relationship between a marketing activity and an online activity;
receiving, by the processor, first aggregate marketing data during a first time period prior to the marketing activity, wherein the first aggregate marketing data includes information about a set of products featured in offline advertising and does not include information about individual users;
identifying, by the processor, a product in the set of products featured in the offline advertising and having online activity data available for the online activity;
selecting, by the processor, a control product based on the control product not being in the set of products featured in the offline advertising and having the online activity data available;
collecting, by the processor, the online activity data for the product and the control product during a time period corresponding to the marketing activity, wherein the online activity data includes the information about the individual users;
determining, by the processor, a prior distribution corresponding to a product effect coefficient based at least in part on the online activity data for the product and the control product;
generating, by the processor using the artificial neural network, a regression model that includes the product effect coefficient based on the prior distribution, wherein the regression model represents the relationship between the marketing activity and the online activity;
estimating, by the processor, a treatment effect for the marketing activity on the online activity based on the regression model;
receiving, by the processor, second aggregate marketing data during a second time period after the marketing activity; and
updating, by the processor, the artificial neural network based on the second aggregate marketing data.