Granted Patent B2

US 11,553,085 · App. 17/111,807 · Granted Jan 10, 2023

Inventor: Sekar Krishnan (Thuckalay, IN)

Assignee: UNIPHORE SOFTWARE SYSTEMS, INC.

H04M3/5175G06N20/00G06Q30/0201G10L15/02G10L15/063G10L15/1815G10L15/22

Abstract

Claims (51)

1. A method for predicting satisfaction of a customer in relation to a call between the customer and an agent, the method comprising:

dividing, at a call analytics server (CAS), a transcribed text of the call into a plurality of phases of a conversation,

extracting, at the CAS, at least one call feature for each of the plurality of phases;

receiving, at the CAS, call metadata;

extracting, at the CAS, metadata features from the call metadata;

combining, at the CAS, the call features and the metadata features; and

generating, using a trained machine learning (ML) model, an output based on the combined features, the output indicating whether the customer is satisfied or not,

wherein the ML model is trained to generate an output indicating whether the customer is satisfied or not, based on an input of the combined features.

2. The method of claim 1 , wherein the plurality of phases comprises an opening phase, a substantive phase, a closing phase, and an overall phase corresponding to each of the entire conversation, the customer conversation, and the agent conversation, yielding a total of 12 phases.

3. The method of claim 2 , wherein the at least one call feature comprises at least one of a total number of words spoken, a sentiment score, positivity or negativity,

wherein the sentiment score is calculated by a sentiment analysis module

wherein the positivity is a percentage of positive words, and

wherein the negativity is a percentage of negative words.

4. The method of claim 3 , wherein the at least one call feature comprises the total number of words spoken, the sentiment score, positivity and negativity for each of the 12 phases, and

wherein the at least one call feature comprises 48 call features.

5. The method of claim 4 , wherein the call metadata comprises information relating to at least one of the customer, the agent, the call, a business with which the call relates to.

6. The method of claim 5 , wherein the metadata features comprise at least one of agent identification, agent department, resolution provided by the agent, customer identification, geographical location of the customer, language spoken by the customer, customer vocation, customer rating, customer feedback, business type, promotion(s) available, call duration, number of holds, duration of each hold, or total duration of hold(s).

7. A method for training a machine learning (ML) model for predicting satisfaction of a customer in relation to a call between the customer and an agent, the method comprising:

for each of a first plurality of training calls, for each of which a customer satisfaction outcome is known, receiving, at a call analytics server (CAS), a transcribed text of a training call;

dividing, at the CAS, the transcribed text into a plurality of phases of a conversation;

extracting, at the CAS, at least one call feature for each of the plurality of phases;

receiving, at the CAS, call metadata for each of the first plurality of training calls;

extracting, at the CAS, metadata features from the call metadata;

combining, at the CAS, the call features and the metadata features for each call of the first plurality of training calls; and

providing the combined features of the training call, and the known outcome of customer satisfaction of the training call to the ML model.

8. The method of claim 7 , further comprising:

after training the ML model using the first plurality of training calls, determining the accuracy of the ML model using a second plurality of training calls; and

if the accuracy of the ML model is below a predefined threshold, training the ML model using a third plurality of training calls.

9. The method of claim 7 , wherein the plurality of phases comprises an opening phase, a substantive phase, a closing phase, and an overall phase corresponding to each of the entire conversation, the customer conversation, and the agent conversation, yielding a total of 12 phases,

wherein at least one call feature comprises the total number of words spoken, the sentiment score, positivity and negativity, and

wherein a total of 48 call features are extracted.

10. The method of claim 9 , wherein the metadata features comprise at least one of agent identification, agent department, resolution provided by the agent, customer identification, geographical location of the customer, language spoken by the customer, customer vocation, customer rating, customer feedback, business type, promotion(s) available, call duration, number of holds, duration of each hold, or total duration of hold(s).

11. An apparatus for predicting satisfaction of a customer in relation to a call between the customer and an agent, the apparatus comprising:

at least one processor;

a memory communicably coupled to the at least one processor, the memory comprising computer executable instructions, which when executed using the at least one processor, perform a method comprising:

dividing, at a call analytics server (CAS), a transcribed text of the call into a plurality of phases of a conversation,

extracting, at the CAS, at least one call feature for each of the plurality of phases,

receiving, at the CAS, call metadata,

extracting, at the CAS, metadata features from the call metadata,

combining, at the CAS, the call features and the metadata features, and

generating, using a trained machine learning (ML) model, an output based on the combined features, the output indicating whether the customer is satisfied or not,

wherein the ML model is trained to generate an output indicating whether the customer is satisfied or not, based on an input of the combined features.

12. The apparatus of claim 11 , wherein the plurality of phases comprises an opening phase, a substantive phase, a closing phase, and an overall phase corresponding to each of the entire conversation, the customer conversation, and the agent conversation, yielding a total of 12 phases.

13. The apparatus of claim 12 , wherein the at least one call feature comprises at least one of a total number of words spoken, a sentiment score, positivity or negativity,

wherein the sentiment score is calculated by a sentiment analysis module,

wherein the positivity is a percentage of positive words, and

wherein the negativity is a percentage of negative words.

14. The apparatus of claim 12 , wherein the at least one call feature comprises the total number of words spoken, the sentiment score, positivity and negativity for each of the 12 phases, and

wherein the at least one call feature comprises 48 call features.

15. The apparatus of claim 13 , wherein the call metadata comprises information relating to at least one of the customer, the agent, the call, a business with which the call relates to.

16. The apparatus of claim 15 , wherein the metadata features comprise at least one of agent identification, agent department, resolution provided by the agent, customer identification, geographical location of the customer, language spoken by the customer, customer vocation, customer rating, customer feedback, business type, promotion(s) available, call duration, number of holds, duration of each hold, or total duration of hold(s).

Assignments (9)

Priority Claims (1)

Continuity (1)