Cynthia J Rogers, 6710399 Beckett Pl, Truckee, CA 96161

2022027, Sep 1, 2022

Feb 26, 2021

17/187737

- San Francisco CA, US
William Robert Pentney - San Francisco CA, US
Eric M Pollmann - Los Altos CA, US
Cynthia E Rogers - Boston MA, US
Mustafa Paksoy - San Francisco CA, US
Zachery Abe Miranda - San Francisco CA, US

Amplitude Inc. - San Francisco CA

G06N 5/04
G06N 20/00
G06N 5/02

Implementations described herein relate to methods, systems, and computer-readable media for automated generation and use of a machine learning (ML) model to provide recommendations. In some implementations, a method includes receiving a recommendation specification that includes a content type and an outcome identifier, and determining model parameters for a ML model based on the recommendation specification. The method further includes generating a historical user feature matrix (FM), generating a historical content feature matrix (FM), and transforming the historical user FM and the historical content FM into a suitable format for the ML model. The method further includes obtaining a target dataset that includes historical results for the outcome identifier for a plurality of pairs of user identifiers and content items of the content type. The method further includes training the ML model using supervised learning to generate a ranked list of content items for each user identifier.

2021003, Feb 4, 2021

Jul 29, 2019

16/525457

- San Francisco CA, US
Cynthia Rogers - San Francisco CA, US
Eric Pollmann - Los Altos CA, US
Muhammad Bilal Mahmood - San Francisco CA, US

G06N 20/00
G06F 17/18
G06F 11/34
G06F 17/16

Systems and methods for computing a causal uplift in performance of an output action for one or more treatment actions in parallel are described herein. In an embodiment, a server computer receives interaction data for a particular period of time which identifies a plurality of users and a plurality of actions that were performed by each user of the plurality of users through a particular graphical user interface during the particular period of time. The server computer uses the interaction data to generate a feature matrix of actions for each user, and a set of confounding variables included to minimize spurious correlations. The feature matrix is then used to train a machine learning system, using data identifying a user's performance or non-performance of each action as inputs and data identifying performance or non-performance of a target output action as the output. A treatment effect is then computed for a treatment action by generating a simulated treatment matrix where all values for the treatment action are set to true, computing an average of outputs from the machine learning system using the simulated treatment matrix, generating a simulated control matrix where all values for the treatment action are set to false, computing an average of outputs from the machine learning system using the simulated control matrix, and computing a difference between the two average outputs.