System and method for score normalization

1 . A computer implemented method for mapping a first set of output values of a first learning model to a distribution of a second set of output values of a second learning model, the method comprising, using a computer processor:

calculating a first distribution function for a set of source values;

calculating a second distribution function for a set of target values;

calculating a first set of quantiles for the first distribution function, and a second set of quantiles for the second distribution function;

using said first and second set of quantiles in a linear interpolation of the set of target values to obtain a source values array and a matched interpolated values array;

calculating an absolute distance from each value in the source values array to the first set of output values of the first learning model;

determining a corresponding value in the matched interpolated values array corresponding to a value in the source values array which has the smallest said absolute distance; and

outputting a set of matched values,

wherein the first and second learning models are one of: the same type, or a different type, and

wherein the type is one of: a convolutional neural network (CNN), a support vector machine (SVM), k-nearest neighbor (KNN), generative additive model (GAM), a decision tree, neural network, or deep neural network.

2 . The computer implemented method of claim 1 , wherein the set of source values and the set of target values comprise a representation of continuous data values, and wherein calculating said first and second distribution functions comprises calculating a first cumulative distribution function (CDF) for a first plurality of histogram scores for the set of source values, and calculating a second CDF for a second plurality of histogram scores for the set of target values.

3 . The computer implemented method of claim 1 , wherein calculating said first and second sets of quantiles comprises dividing the first distribution function by the size of the set of source values and dividing the second distribution function by the size of the set of target values.

4 . The computer implemented method of claim 1 , wherein the first and second set of output values of the first and second learning model comprise prediction scores for one or more monitored variables monitored based on natural language processing (NLP) of at least one of a transcript, or an audio recording, of an interaction between an agent and a customer.

5 . The computer implemented method of claim 4 , wherein the interaction is one of: a phone conversation, one or a plurality of emails, one or a plurality of text messages, one or a plurality of social media comments, one or a plurality of web chat messages, an audio transcription, or a textual representation of a digital channel.

6 . The computer implemented method of claim 4 , wherein the set of matched values is used in automatically determining agent performance based on one or more thresholds defined for the distribution of the second set of output values of the second learning model.

7 . The computer implemented method of claim 1 , wherein the first and second set of output values of the first and second learning models comprise prediction scores for one or more monitored variables monitored based on natural language processing (NLP) of at least one of a transcript, or an audio recording, of an interaction between a customer and an interactive voice response (IVR), and

wherein the set of matched values is used in automatically routing a call based on one or more thresholds defined for the distribution of the second set of output values of the second learning model.

8 . A system for mapping a first set of output values of a first learning model to a distribution of a second set of output values of a second learning model, the system comprising:

at least one processor; and

a memory containing instructions which, when executed by the at least one processor, cause the at least one processor to:

calculate a first distribution function for a set of source values;

calculate a second distribution function for a set of target values;

calculate a first set of quantiles for the first distribution function, and a second set of quantiles for the second distribution function;

use said first and second set of quantiles in a linear interpolation of the set of target values to obtain a source values array and a matched interpolated values array;

calculate an absolute distance from each value in the source values array to the first set of output values of the first learning model;

determine a corresponding value in the matched interpolated values array corresponding to a value in the source values array which has the smallest said absolute distance; and

output a set of matched values,

wherein the first and second learning models are one of: the same type, or a different type, and

wherein the type is one of: a convolutional neural network (CNN), a support vector machine (SVM), k-nearest neighbor (KNN), generative additive model (GAM), a decision tree, neural network, or deep neural network.

9 . The system of claim 8 , wherein the set of source values and the set of target values comprise a representation of continuous data values, and wherein the processor is configured to calculate said first and second distribution functions by calculating a first cumulative distribution function (CDF) for a first plurality of histogram scores for the set of source values, and calculate a second CDF for a second plurality of histogram scores for the set of target values.

10 . The system of claim 8 , wherein the at least one processor is configured to calculate said first and second sets of quantiles by dividing the first distribution function by the size of the set of source values and by dividing the second distribution function by the size of the set of target values.

11 . The system of claim 8 , wherein the first and second set of output values of the first and second learning models comprise prediction scores for one or more monitored variables monitored based on natural language processing (NLP) of at least one of a transcript, or an audio recording, of an interaction between an agent and a customer.

12 . The system of claim 11 , wherein the interaction is one of: a phone conversation, one or a plurality of emails, one or a plurality of text messages, one or a plurality of social media comments, one or a plurality of web chat messages, an audio transcription, or a textual representation of a digital channel.

13 . The system of claim 11 , wherein the set of matched values is used in automatically determining agent performance based on one or more thresholds defined for the distribution of the second set of output values of the second learning model.

14 . The system of claim 8 , wherein the first and second set of output values of the first and second learning models comprise prediction scores for one or more monitored variables monitored based on natural language processing (NLP) of at least one of a transcript, or an audio recording, of an interaction between a customer and an interactive voice response (IVR), and

wherein the at least one processor is configured to automatically use the set of matched values to automatically route a call based on one or more thresholds defined for the distribution of the set of output values of the second learning model.