Methods and systems for customer churn prediction

• US 10,580,020 B2
• Filed: 07/28/2015
• Issued: 03/03/2020
• Est. Priority Date: 07/28/2015
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Inventors: Sharma, Avinash, Annamaneni, Narendra, Gupta, Divanshu, Manjunath, Geetha
• Primary Examiner(s): Goldberg, Ivan R
• Application Number:
  US14/810,508
• Publication Number:
  US 20170032391A1
• Time in Patent Office: 1,680 Days
• Field of Search: None
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities
• CPC Class Codes: G06Q 30/0204 Market segmentation
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

A method and a system are provided for customer churn prediction. The method includes creating a graph comprising a plurality of nodes and a plurality of edges. At least one edge of the plurality of edges in the graph connects more than two nodes of the plurality of nodes. Each of the plurality of nodes represents a customer. Thereafter, a similarity matrix representative of a similarity between each of a plurality of customers is determined based on the graph. Further, the similarity matrix is decomposed into a first matrix and a second matrix. A third matrix is determined based on the first matrix, the second matrix, and a scaling parameter. The third matrix is utilized to identify a set of potential churn customers from the plurality of customers. Thereafter, the set of potential churn customers is presented on a user-computing device.

Claims

1. A method of operating a customer service system, the method comprising:

receiving, by a database server, customer data for a plurality of customers through a communication network from a user-computing device, the plurality of customers comprising a set of churner customers and a set of non-churner customers;

fusing, by the database server, the customer data;

storing, by the database server, the fused customer data;

receiving, by the database server, a query from an application server through the communication network to extract the fused customer data, the application server comprising one or more processors;

receiving, by the application server, the extracted customer data comprising a plurality of attributes through the communication network;

defining, by the application server, a plurality of bins corresponding to the plurality of attributes using a density plot of numerical values of each attribute of the plurality of attributes;

generating, by the application server, an incidence matrix, based on the plurality of attributes, comprising a plurality of rows and a plurality of columns, wherein each row of the plurality of rows represents a customer of the plurality of customers and the plurality of columns corresponds to the plurality of bins;

creating, by the application server, a customer similarity graph comprising a plurality of customer nodes and a plurality of edges, wherein each edge represents a bin of the plurality of bins, at least one edge of the plurality of edges connects more than two customer nodes of the plurality of customer nodes, each of the plurality of customer nodes represents a customer, the plurality of customer nodes are located in graphical neighborhoods, the customer similarity graph is created based on the incidence matrix, and each edge of the plurality of edges is represented as an n-dimensional vector with binary entries;

determining, by the application server, a similarity matrix representative of a similarity between each of the plurality of customers, represented by the plurality of customer nodes, based on the customer similarity graph;

decomposing, by the application server, the similarity matrix into a first matrix and a second matrix;

determining, by the application server, a third matrix based on the first matrix, the second matrix, and a scaling parameter, wherein the third matrix is a heat kernel matrix defining an effect of each of the plurality of customers on the other customers of the plurality of customers, and the scaling parameter represents a factor that defines a radius of a neighborhood to determine an influence of the customer propagated among the plurality of customers;

receiving, by the application server, information comprising a first set of attributes from the plurality of attributes, the first set of attributes pertaining to a set of churner customers from the plurality of customers, who have left the service during a pre-defined time interval through the communication network from the user-computing device, and a second set of attributes from the plurality of attributes pertaining to a set of non-churner customers from the plurality of customers;

determining, by the application server, a set of potential churn customers based on;

the third matrix comprising the effect of the customer on the other customers in the radius of the neighborhood and the influence of the customer that propagated, and

the information comprising the first set of attributes pertaining to the set of churner customers;

receiving, by the user-computing device from the application server through the communication network, the set of potential churn customers; and

visualizing, through a graphical interface on the user-computing device, the set of potential churn customers on the customer similarity graph, the plurality of customer nodes comprising churner customer nodes, non-churner customer nodes, and potential churn customer nodes located in the graphical neighborhoods relative to the radius, the potential churn customers comprising a third set of attributes from the plurality of attributes, the third set of attributes pertaining to the set of potential churn customers and based on;

the first set of attributes of the churner customers,

the second set of attributes of the non-churner customers,

the scaling parameter with respect to the potential churn customers relative the churner customers, and

the scaling parameter with respect to the potential churn customers relative to the non-churner customers,

wherein the visualization configured to assist a user in retaining the set of potential churn customers.

2. The method of claim 1, wherein determining the set of potential churn customers comprises:

creating, by the application server, a fourth matrix, comprising a first value or a second value assigned to the plurality of customers, based on the information pertaining to the set of churner customers, the first value being assigned to the set of non-churner customers, and the second value being assigned to the set of churner customers;

determining, by the application server, a customer influence value associated with each customer in the plurality of customers based on the third matrix and the fourth matrix; and

comparing, by the application server, the customer influence value associated with each customer in the plurality of customers with a pre-defined threshold to identify the set of potential churn customers from the plurality of customers.

3. The method of claim 1, further comprising:

receiving, by the application server, an updated fourth matrix, from the user-computing device; and

updating the set of potential churn customers based on the updated fourth matrix.

4. The method of claim 1, further comprising assigning, by the application server, a weight to an edge, from the plurality of edges, wherein the weight represents a measure of similarity between a set of customers, from the plurality of customers, encompassed by the edge in the graph;

wherein the weight assigned to the edge is less in comparison to a weight assigned to each of remaining edges in the plurality of edges, when a number of nodes encompassed by the edge is greater than a pre-defined threshold.

5. The method of claim 1, wherein the plurality of attributes comprises at least one of a customer gender, a customer geography, a customer age, a customer annual billing amount, a customer service type, a customer industry type, a customer voice usage, a customer data usage, a number of years of a customer with association with a bank, a customer account type, a customer credit limit, and a customer annual spending amount.

6. The method of claim 1, further comprising generating, by the application server, a histogram indicative of the density plot associated with each of the plurality of attributes received.

7. The method of claim 1, wherein the density plot represents a relationship between a number of customers of the plurality of customers and an attribute of the plurality of attributes, each bin of the plurality of bins has an equal number of customers, and the number of bins is greater in areas of higher density of the density plot.

8. The method of claim 1, wherein the customer similarity graph comprising training nodes and testing node, the training nodes comprising the churner customer nodes and the non-churner customer nodes, and the testing nodes comprising the potential churn customer nodes.

9. A customer service system comprising:

a database server configured to;

receive customer data for a plurality of customers through a communication network from a user-computing device, the plurality of customers comprising a set of churner customers and a set of non-churner customers,

fuse the customer data,

store the fused customer data;

receive a query from an application server through the communication network to extract the fused customer data, the application server comprising one or more processors;

an application server configured to;

receive the extracted customer data comprising a plurality of attributes through the communication network,

define a plurality of bins corresponding to the plurality of attributes using a density plot of numerical values of each attribute of the plurality of attributes,

generate an incidence matrix, based on the plurality of attributes, comprising a plurality of rows and a plurality of columns, wherein each row of the plurality of rows represents a customer of the plurality of customers and the plurality of columns corresponds to the plurality of bins,

create a customer similarity graph that comprises a plurality of customer nodes and a plurality of edges, wherein each edge represents a bin of the plurality of bins, at least one edge of the plurality of edges connects more than two customer nodes of the plurality of customer nodes, each of the plurality of customer nodes represents a customer, the plurality of customer nodes are located in graphical neighborhoods, the customer similarity graph is created based on the incidence matrix, each edge of the plurality of edges is represented as an n-dimensional vector with binary entries, and the plurality of edges represents a plurality of attributes associated with the plurality of customers,

determine a similarity matrix representative of a similarity between each of a plurality of customers, represented by the plurality of customer nodes, based on the customer similarity graph,

decompose the similarity matrix into a first matrix and a second matrix,

determine a third matrix based on the first matrix, the second matrix, and a scaling parameter, wherein the third matrix is a heat kernel matrix defining an effect of each of the plurality of customers on the other customers of the plurality of customers, and the scaling parameter represents a factor that defines a radius of neighborhood to determine an influence of the customer propagated among the plurality of customers;

receive information comprising a first set of attributes from the plurality of attributes, the first set of attributes pertaining to a set of churner customers from the plurality of customers, who have left the service during a pre-defined time interval through the communication network from the user-computing device, and a second set of attributes from the plurality of attributes pertaining to a set of non-churner customers from the plurality of customers,

determine a set of potential churn customers based on;

the third matrix comprising the effect of the customer on the other customers in the radius of the neighborhood and the influence of the customer that propagated, and

the information comprising the first set of attributes pertaining to the set of churner customers; and

a user computing device configured to;

receive from the application server through the communication network, the set of potential churn customers; and

visualize the set of potential churn customers through a graphical interface on the user-computing device on the customer similarity graph, the plurality of customer nodes comprising churner customer nodes, non-churner customer nodes, and potential churn customer nodes located in the graphical neighborhoods relative to the radius, the potential churn customers comprising a third set of attributes from the plurality of attributes, the third set of attributes pertaining to the set of potential churn customers and based on;

the first set of attributes of the churner customers,

the second set of attributes of the non-churner customers,

the scaling parameter with respect to the potential churn customers relative the churner customers, and

the scaling parameter with respect to the potential churn customers relative to the non-churner customers,

the visualization configured to assist a user in retaining the set of potential churn customers.

10. The application server of claim 9, wherein the application server is further configured to determine the set of potential churn customers by:

creating a fourth matrix comprising a first value or a second value assigned to the plurality of customers, based on the set of churner customers, the first value being assigned to the set of non-churner customers, and the second value being assigned to the set of churner customers;

determining, by the application server, a customer influence value associated with each customer in the plurality of customers based on the third matrix and the fourth matrix; and

comparing, by the application server, the customer influence value associated with each customer in the plurality of customers with a pre-defined threshold to identify the set of potential churn customers from the plurality of customers.

11. The application server of claim 10, wherein the application server is further configured to:

receive an updated fourth matrix from the user-computing device; and

update the set of potential churn customers based on the updated fourth matrix.

12. The application server of claim 10, wherein the application server is further configured to assign a weight to an edge, from the plurality of edges, wherein the weight represents a measure of similarity between a set of customers, from the plurality of customers, encompassed by the edge in the graph, and the weight assigned to the edge is less in comparison to a weight assigned to each of remaining edges in the plurality of edges, when a number of nodes encompassed by the edge is greater than a pre-defined threshold.

13. The system of claim 9, wherein the plurality of attributes comprises at least one of a customer gender, a customer geography, a customer age, a customer annual billing amount, a customer service type, a customer industry type, a customer voice usage, a customer data usage, a number of years of a customer with association with a bank, a customer account type, a customer credit limit, and a customer annual spending amount.

14. The system of claim 9, wherein the application server is further configured to generate a histogram indicative of the density plot associated with each of the plurality of attributes received.

15. The system of claim 9, wherein the density plot represents a relationship between a number of customers of the plurality of customers and an attribute of the plurality of attributes, each bin of the plurality of bins has an equal number of customers, and the number of bins is greater in areas of higher density of the density plot.

16. The system of claim 9, wherein the customer similarity graph comprising training nodes and testing node, the training nodes comprising the churner customer nodes and the non-churner customer nodes, and the testing nodes comprising the potential churn customer nodes.

17. A non-transitory computer-readable storage medium having stored thereon, a set of computer-executable instructions for causing a customer service system, the system comprising a database server, a communication network, a user-computing device, and an application server comprising one or more processors, to perform steps comprising:

receiving, by a database server, customer data for a plurality of customers through a communication network from a user-computing device, the plurality of customers comprising a set of churner customers and a set of non-churner customers;

fusing, by the database server, the customer data;

storing, by the database server, the fused customer data;

receiving, by the database server, a query from an application server through the communication network to extract the fused customer data, the application server comprising one or more processors;

receiving, by the application server, the extracted customer data comprising a plurality of attributes through the communication network;

defining, by the application server, a plurality of bins corresponding to the plurality of attributes using a density plot of numerical values of each attribute of the plurality of attributes;

generating, by the application server, an incidence matrix, based on the plurality of attributes, comprising a plurality of rows and a plurality of columns, wherein each row of the plurality of rows represents a customer of the plurality of customers and the plurality of columns corresponds to the plurality of bins;

creating, by the application server, a customer similarity graph comprising a plurality of customer nodes and a plurality of edges, wherein each edge represents a bin of the plurality of bins, at least one edge of the plurality of edges connects more than two customer nodes of the plurality of customer nodes, each of the plurality of customer nodes represents a customer, the plurality of customer nodes are located in graphical neighborhoods, the customer similarity graph is created based on the incidence matrix, each edge of the plurality of edges is represented as an n-dimensional vector with binary entries, and the plurality of edges represents a plurality of attributes associated with the plurality of customers;

determining, by the application server, a similarity matrix representative of a similarity between each of a plurality of customers, represented by the plurality of customer nodes, based on the customer similarity graph;

decomposing, by the application server, the similarity matrix into a first matrix and a second matrix;

determining, by the application server, a third matrix based on the first matrix, the second matrix, and a scaling parameter, wherein the third matrix is a heat kernel matrix defining an effect of each of the plurality of customers on the other customers of the plurality of customers, and the scaling parameter represents a factor that defines a radius of neighborhood to determine an influence of the customer propagated among the plurality of customers;

receiving, by the application server, information comprising a first set of attributes from the plurality of attributes, the first set of attributes pertaining to a set of churner customers from the plurality of customers, who have left a service associated with an organization during a pre-defined time interval, and a second set of attributes from the plurality of attributes pertaining to a set of non-churner customers from the plurality of customers;

determining, by the application server, a set of potential churn customers based on;

the third matrix comprising the effect of the customer on the other customers in the radius of the neighborhood and the influence of the customer that propagated, and

the information comprising the first set of attributes pertaining to the set of churner customers;

receiving, by the user-computing device from the application server through the communication network, the set of potential churn customers; and

visualizing the set of potential churn customers through a graphical interface on user-computing device on the customer similarity graph, the plurality of customer nodes comprising churner customer nodes, non-churner customer nodes, and potential churn customer nodes located in the graphical neighborhoods relative to the radius, the potential churn customers comprising a third set of attributes from the plurality of attributes, the third set of attributes pertaining to the set of potential churn customers and based on;

the first set of attributes of the churner customers,

the second set of attributes of the non-churner customers,

the scaling parameter with respect to the potential churn customers relative the churner customers, and

the scaling parameter with respect to the potential churn customers relative to the non-churner customers,

the visualization configured to assist a user in retaining the set of potential churn customers.

18. The non-transitory computer-readable storage medium of claim 17, wherein the plurality of attributes comprises at least one of a customer gender, a customer geography, a customer age, a customer annual billing amount, a customer service type, a customer industry type, a customer voice usage, a customer data usage, a number of years of a customer with association with a bank, a customer account type, a customer credit limit, and a customer annual spending amount.

19. The non-transitory computer-readable storage medium of claim 17, wherein determining the set of potential churn customers further comprises:

creating, by the application server, a fourth matrix, comprising of a first value or a second value assigned to the plurality of customers, based on the information pertaining to the set of churner customers, the first value being assigned to the set of non-churner customers, and the second value being assigned to the set of churner customers;

determining, by the application server, a customer influence value associated with each customer in the plurality of customers based on the third matrix and the fourth matrix; and

comparing, by the application server, the customer influence value associated with each customer in the plurality of customers with a pre-defined threshold to identify the set of potential churn customers from the plurality of customers.

20. The non-transitory computer-readable storage medium of claim 17, wherein the steps further comprise generating, by the application server, a histogram indicative of the density plot associated with each of the plurality of attributes received.

21. The non-transitory computer-readable storage medium of claim 17, wherein the density plot represents a relationship between a number of customers of the plurality of customers and an attribute of the plurality of attributes, each bin of the plurality of bins has an equal number of customers, and the number of bins is greater in areas of higher density of the density plot.

22. The non-transitory computer-readable storage medium of claim 17, wherein the customer similarity graph comprising training nodes and testing node, the training nodes comprising the churner customer nodes and the non-churner customer nodes, and the testing nodes comprising the potential churn customer nodes.