US 12,260,422 B2
Generation of models for classifying user groups
Nicole Woytarowicz, Allen, TX (US); Samuel Vaughn Tucker, Minneapolis, MN (US); Inga Mgherbrishvili, Kirkland, WA (US); Rowan Michael Wing, Longmont, CO (US); and Hanhan Wang, Seattle, WA (US)
Assigned to Twilio Inc., San Francisco, CA (US)
Filed by Twilio Inc., San Francisco, CA (US)
Filed on Mar. 21, 2022, as Appl. No. 17/655,724.
Prior Publication US 2023/0298055 A1, Sep. 21, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06Q 30/0204 (2023.01); G06N 20/00 (2019.01); G06Q 30/0202 (2023.01)
CPC G06Q 30/0204 (2013.01) [G06N 20/00 (2019.01); G06Q 30/0202 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
generating, by one or more processors, training data that comprises values for features that include event features, user information features, audience labels, and product features, the generating of the training data including labelling each event among events within a first time window, the labelling of the events distinguishing first events associated with an audience category from second events not associated with the audience category;
training, by the one or more processors, a propensity-to-buy machine-learning (ML) model based on the generated training data that comprises the labelled events and the values for the features that include the event features, the user information features, the audience labels, and the product features;
accessing, by the one or more processors, further events, each further event comprising a data structure describing an operation performed by a single persona identified among a plurality of users;
providing, by the one or more processors, event information of the further events and information of the single persona as input to the propensity-to-buy ML model trained based on the training data that comprises the labelled events and the values for the features that include the event features, the user information features, the audience labels, and the product features;
generating, by the propensity-to-buy ML model, a score for the single persona indicating a probability that the single persona belongs to the audience category and will purchase a product within a future time window;
determining, by the one or more processors, a total value of billing events for the single persona over the future time window based on the score;
updating, by the one or more processors, the training data based on at least one of a new event feature or an obsolete event feature, the updating of the training data including labelling each event among events within a second time window, the labelling of the events distinguishing third events associated with the audience category from fourth events not associated with the audience category; and
retraining, by the one or more processors, the propensity-to-buy ML model based on the updated training data.