Customer service ticket evaluation using multiple time-based machine learning models customer

Patent metadata

Field	Value
Publication number	US-11587094-B2
Application number	US-201916587364-A
Country	US
Kind code	B2
Filing date	Sep 30, 2019
Priority date	Sep 30, 2019
Publication date	Feb 21, 2023
Grant date	Feb 21, 2023

Abstract

Techniques are provided for customer service ticket prioritization using multiple time-based machine learning models. One method comprises obtaining a customer service ticket; collecting, in response to the obtaining, features related to the customer service ticket comprising dynamic information regarding activities performed to address the customer service ticket and textual information describing the customer service ticket; applying the collected features to one of multiple machine learning models to obtain a distress score indicating a likelihood that the customer service ticket will become a distressed customer service ticket, wherein the multiple machine learning models each correspond to different time periods and are trained on historical data and the collected features are applied to the one machine learning model based on an age of the customer service ticket; and processing multiple customer support tickets based on the distress score. The machine learning models are trained using cumulative corresponding historical training data associated with a respective one of hourly, daily, weekly and monthly time periods.

First claim

What is claimed is:

1. A method, comprising:
obtaining a given customer service ticket;
collecting, in response to the obtaining, a plurality of features related to the given customer service ticket, wherein the plurality of features comprises dynamic information regarding activities performed to address the given customer service ticket and textual information describing at least a portion of the given customer service ticket;
obtaining a plurality of machine learning models, wherein each of the plurality of machine learning models processes customer service tickets of a different ticket age, wherein each respective machine learning model is trained on a corresponding set of training data comprising features of historical customer support tickets having a ticket age that corresponds to the ticket age processed by the respective machine learning model, wherein the training data for at least a first one of the machine learning models comprises at least a portion of the training data for at least a second one of the machine learning models, and wherein the second machine learning model processes customer service tickets having a younger ticket age than the customer service tickets processed by the first machine learning model;
selecting a given one of the plurality of machine learning models based at least in part on a ticket age of the given customer support ticket relative to the respective ticket ages of the customer service tickets processed by the plurality of machine learning models;
applying the collected plurality of features related to the given customer service ticket to the selected machine learning model to obtain a distress score indicating a likelihood that the given customer service ticket will become a distressed customer service ticket; and
processing at least the given customer support ticket based at least in part on the distress score;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.

2. The method of claim 1 , wherein the plurality of features related to the given customer service ticket further comprise one or more of a frequency of activities performed to address the given customer service ticket, product information, customer information and install base information.

3. The method of claim 1 , wherein the textual information describing the given customer service ticket captures a theme of the given customer service ticket.

4. The method of claim 1 , wherein the textual information describing the given customer service ticket is extracted from the given customer service ticket based on one or more of a term frequency, a topic model and a vector representation.

5. The method of claim 1 , wherein the plurality of machine learning models processing customer service tickets of a different ticket age each correspond to at least one of: (i) a different number of hours since the initiation of the given customer support ticket, (ii) a different number of days since the initiation of the given customer support ticket, (iii) a different number of weeks since the initiation of the given customer support ticket, and (iv) a different number of months since the initiation of the given customer support ticket.

6. The method of claim 5 , wherein each of the plurality of machine learning models processing customer service tickets of the different ticket age is trained using corresponding training data associated with the at least one of: (i) the different number of hours since the initiation of the given customer support ticket, (ii) the different number of days since the initiation of the given customer support ticket, (iii) the different number of weeks since the initiation of the given customer support ticket, and (iv) the different number of months since the initiation of the given customer support ticket.

7. The method of claim 6 , wherein the corresponding training data associated with a given machine learning model comprises cumulative training data based at least in part on the historical customer support tickets having a ticket age up to the ticket age associated with given machine learning model.

8. The method of claim 1 , wherein each of the plurality of machine learning models is based on random forest models.

9. The method of claim 1 , wherein the plurality of machine learning models comprises a base machine learning model used for a time period immediately following the obtaining, wherein the base machine learning model processes one or more features comprising a product group, a support ticket severity, an asset age, one or more customer-related historical features and one or more product-related historical features.

10. A computer program product, comprising a tangible machine-readable storage medium having encoded therein executable code of one or more software programs, wherein the one or more software programs when executed by at least one processing device perform the following steps:
obtaining a given customer service ticket;
collecting, in response to the obtaining, a plurality of features related to the given customer service ticket, wherein the plurality of features comprises dynamic information regarding activities performed to address the given customer service ticket and textual information describing at least a portion of the given customer service ticket;
obtaining a plurality of machine learning models, wherein each of the plurality of machine learning models processes customer service tickets of a different ticket age, wherein each respective machine learning model is trained on a corresponding set of training data comprising features of historical customer support tickets having a ticket age that corresponds to the ticket age processed by the respective machine learning model, wherein the training data for at least a first one of the machine learning models comprises at least a portion of the training data for at least a second one of the machine learning models, and wherein the second machine learning model processes customer service tickets having a younger ticket age than the customer service tickets processed by the first machine learning model;
selecting a given one of the plurality of machine learning models based at least in part on a ticket age of the given customer support ticket relative to the respective ticket ages of the customer service tickets processed by the plurality of machine learning models;
applying the collected plurality of features related to the given customer service ticket to the selected machine learning model to obtain a distress score indicating a likelihood that the given customer service ticket will become a distressed customer service ticket; and
processing at least the given customer support ticket based at least in part on the distress score.

11. The computer program product of claim 10 , wherein the textual information describing the customer service ticket captures a theme of the given customer service ticket.

12. The computer program product of claim 10 , wherein the plurality of machine learning models processing customer service tickets of a different ticket age each correspond to at least one of: (i) a different number of hours since the initiation of the given customer support ticket, (ii) a different number of days since the initiation of the given customer support ticket, (iii) a different number of weeks since the initiation of the given customer support ticket, and (iv) a different number of months since the initiation of the given customer support ticket, and wherein each of the plurality of machine learning models processing customer service tickets of the different ticket age is trained using corresponding training data associated with th

Assignees

• Emc Ip Holding Co Llc

Inventors

• Tzur Anat Parush
• Savir Amihai

Classifications

• G06Q30/016 Primary

  After-sales · CPC title

• G06N20/20

  Ensemble learning · CPC title