US 12,335,276 B2
Exponentially smoothed categorical encoding to control access to a network resource
Nathan Daniel Monnig, Boise, ID (US); Andrew Nader Rafla, Boise, ID (US); and Samuel Ward Schrader, Boise, ID (US)
Assigned to KOUNT INC., Boise, ID (US)
Filed by Kount Inc., Boise, ID (US)
Filed on Dec. 31, 2021, as Appl. No. 17/646,696.
Prior Publication US 2023/0216866 A1, Jul. 6, 2023
Int. Cl. H04L 9/40 (2022.01)
CPC H04L 63/1416 (2013.01) [H04L 63/102 (2013.01); H04L 63/1425 (2013.01); H04L 63/1466 (2013.01); H04L 63/20 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
detecting historical events occurring over a network, wherein at least one of the historical events is associated with an observed value of a categorical variable;
updating, through a recursive technique, a numerical aggregate value representing the observed value of the categorical variable, the numerical aggregate value based on both (i) an exponentially smoothed prior numerical aggregate value representing prior historical events associated with the observed value and (ii) an exponentially smoothed positive count aggregate representing a number of times the historical events associated with the observed value led to an outcome of interest, wherein the numerical aggregate value is computed as an exponentially decayed event function and a decay parameter controls a rate at which contribution of historical events to the numerical aggregate value decays as time passes;
detecting an event occurring over the network and associated with the observed value of the categorical variable;
extracting features from the event, wherein the features comprise an encoded feature based on the numerical aggregate value to represent the observed value of the categorical variable;
applying a predictive model, the predictive model having been trained on at least the numerical aggregate value, to the features to determine a score representing likelihood of an outcome for the event; and
based on the score, controlling access to a resource of the network.