FIELD OF THE INVENTION
The present invention relates to churn of network subscribers.
BACKGROUND OF THE INVENTION
Subscribers in mobile networks churn, i.e. unsubscribe from a network operator to
switch to another network operator, for a variety of reasons. For example,
subscribers may churn due to dissatisfaction with services offered by the network
operator (e.g. voice service, data service, video service, short message service
(SMS), multimedia message service (MMS) and the like), dissatisfaction with
service quality, availability of mobile devices on the network, or the like.
Subscribers may also decide to churn from the network operator due to financial
considerations such as competitive pricing, discounts and/or promotions offered
by another network operator.
SUMMARY
According to an embodiment, a system for generating a subscriber churn
prediction comprises a data input device and at least one processor connected to
the data input device. The data input device may receive call detail records from
a network operator detailing communication for at least one subscriber of the
network operator. The at least one processor may execute a churn prediction
program to generate the subscriber churn prediction based at least on the call
detail records.
According to an embodiment, the churn prediction program may include a social
network analysis module that derives social metrics from the call detail records.
According to an embodiment, the churn prediction program may include a churn
prediction module that receives the social metrics from the social network analysis
module and generates the churn prediction.
According to an embodiment, the churn prediction is generated from the social
metrics in combination with at least one traditional metric.
According to an embodiment, the social network analysis module may derive at
least one social metric by determining tie-strengths between connected
subscribers based on one or more calling attributes of the call detail records.
According to an embodiment, the at least one social metric may include at least
one of a net churner influence, a number of neighboring subscribers that are
churners, a number of hops to a nearest churner, a number of calls to churners, a
number of calls to the nearest churner, and a time spent on calls to churners.
According to an embodiment, the social network analysis module may propagate
a churner influence between subscribers based on the tie-strength and determine
a net influence for each subscriber.
According to an embodiment, propagation of the churner influence may be
receiver centric. The influence received by each subscriber from a connected
subscriber may be proportional to the tie-strength of the tie between the
subscribers.
According to an embodiment, a computerized method for generating a prediction
of subscriber churn comprises the steps of receiving, by a churn prediction
program executing on a computer processor, call detail records detailing
communication for at least one subscriber of a network operator and generating,
by the churn prediction program executing on the computer processor, a
subscriber churn prediction based at least on the call detail records.
According to an embodiment, the computerized method may also comprise the
step of deriving, by the churn prediction program executing on the computer
processor, social metrics from the call detail records.
According to an embodiment, the step of generating, by the churn prediction
program executing on the computer processor, the subscriber churn prediction
may include combining the social metrics with at least one traditional metric for
churn prediction.
According to an embodiment, deriving the social metrics may include determining,
by the churn prediction program executing on the computer processor, tiestrengths
between connected subscribers based on one or more calling attributes
of the call detail records.
According to an embodiment, the social metrics may include at least one of a net
churner influence, a number of neighboring subscribers that are churners, a
number of hops to a nearest churner, a number of calls to churners, a number of
calls to the nearest churner, and a time spent on calls to churners.
According to an embodiment, the computerized method may also comprise the
step of propagating, by the churn prediction program executing on the computer
processor, a churner influence between subscribers based on the tie-strength and
determining a net influence for each subscriber.
According to an embodiment, propagating, by the churn prediction program
executing on the computer processor, the churner influence between subscribers
may be receiver centric. The influence received by each subscriber from a
connected subscriber may be proportional to the tie-strength of the tie between
the subscribers.
According to an embodiment, a non-transitory, tangible computer-readable
medium storing instructions adapted to be executed by a computer processor to
perform a method for generating a subscriber churn prediction may comprise the
steps of receiving, by a churn prediction program executing on the computer
processor, call detail records detailing communication for at least one subscriber
of the network operator. The method may also comprise generating, by the churn
prediction program executing on the computer processor, a subscriber churn
prediction based at least on the call detail records.
According to an embodiment, the method may further comprise the step of
deriving, by the churn prediction program executing on the computer processor,
social metrics from the call detail records.
According to an embodiment, the step of generating, by the churn prediction
program executing on the computer processor, the subscriber churn prediction
may include combining the social metrics with at least one traditional metric for
churn prediction.
According to an embodiment, deriving the social metrics may include determining,
by the churn prediction program executing on the computer processor, tiestrengths
between connected subscribers based on one or more calling attributes
of the call detail records.
According to an embodiment, the method may further comprise the step of
propagating, by the churn prediction program executing on the computer
processor, a churner influence between subscribers based on the tie-strength and
determining a net influence for each subscriber.
These and other embodiments of will become apparent in light of the following
detailed description herein, with reference to the accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a schematic diagram of a computerized system according to an
embodiment;
FIG. 2 is a graphical representation of subscriber connections formed by the
computerized system of FIG. 1;
FIG. 3 is a flow diagram of an embodiment for generating churn predictions from
the computerized system of FIG. 1;
FIG. 4 is a graphical representation of an exemplary embodiment of a social circle
formed by the computerized system of FIG. 1; and
FIG. 5 is a schematic diagram of an exemplary embodiment of the computerized
system of FIG. 1.
DETAILED DESCRIPTION
Referring to FIG. 1, an embodiment of a computerized system 10 adapted to
generate a subscriber churn prediction 12 using call detail records 14 from a
network operator 16 is shown. The computerized system 10 includes a social
network analysis module 18 and a churn prediction module 20. The social
network analysis module 18 uses the call detail records 14 to estimate social
network connections between one or more subscribers 22 of the network operator
16. The computerized system 10 inputs churn data 24, which may be provided by
the network operator 16, and the churn prediction module 20 generates the churn
prediction 12 based, at least in part, on the churn data 24 and the estimated social
network connections.
Each call detail record 14 includes one or more calling attributes 25 that detail a
communication between one or more subscribers 22 of the network operator 16.
These call detail records 14 are typically generated and used by the network
operator 16 for billing purposes and the like. The calling attributes 25 of the call
detail records 14 may include, for example, caller and callee identifiers, call date,
call time, call duration, and the location of the cell towers for call initiation and
termination among other details. The call detail records 14 may also include
message sender and recipient identifiers, a message type (e.g. short message
service (SMS) or multimedia message service (MMS)) and a message size.
As discussed above, the social network analysis module 18 of the computerized
system 10 may use the information provided in the call detail records 14 to
estimate strengths of the social network connections between the one or more
subscribers 22 of the network operator 16. For example, referring to FIG. 2, the
social network analysis module 18, shown in FIG. 1, may generate tie-strengths
26 between the subscribers 22 based on the calling attributes 25, shown in FIG.
1, of the call detail records 14 , shown in FIG. 1. Each tie-strength 26 quantifies
the social connection between two subscribers 22 of the network operator 16,
shown in FIG. 1, based on the communication information, i.e. the calling
attributes 25, shown in FIG. 1, provided in the call detail record 14, shown in FIG.
1. For example, the tie-strength 26 between two subscribers 22 may be based on
the number of calls/messages between the subscribers 22, the duration of the
calls, the size of messages, a direction of the calls/messages (e.g. whether or not
the calls/messages were reciprocal), a time to reciprocate calls/messages, the
day/time of calls/messages, a number of common connections with other
subscribers 22 and/or any other similar calling attributes 25, shown in FIG. 1,
between the two subscribers 22. When generating the tie-strengths 26 between
subscribers 22, the social network analysis module 18, shown in FIG. 1,
preferably considers a plurality of the calling attributes 25 since there may be
instances where one calling attribute 25, shown in FIG. 1, alone, may not provide
a good indication of the social connection between two subscribers 22.
Referring to FIG. 3, in operation, the social network analysis module 18, shown in
FIG. 1, may generate the tie-strengths 26 between subscribers 22 by first
normalizing the calling attributes 25 between each of the subscribers 22 at step
28. This allows the calling attributes 25, which are typically measured using
different scales, to be combined by the social network analysis module 18, shown
in FIG. 1, in a non-dimensional manner. For example, call duration may be
measured in minutes while call frequency may be measured in calls per month.
Thus, to make the calling attributes 25 more easily combinable, the social network
analysis module 18 may normalize the calling attributes 25 by rescaling each
calling attribute 25 to have a unit length. For example, in an embodiment, the
social network analysis module 18, shown in FIG. 1, may normalize the calling
attributes 25 by dividing each observation of an attribute x by LI, where:
This operates to rescale each attribute x to have a unit length. Once the calling
attributes 25 have been normalized at step 28, the social network analysis module
18, shown in FIG. 1, may then calculate a weighted sum x of the normalized
attributes x between each of the subscribers 22 at step 30 to provide a measure
of the total calling attributes 25 connecting any two subscribers 22. For example,
the social network analysis module 18, shown in FIG. 1, may calculate the
weighted sum x using the equation:
x = x +c x +...+ c nxn;
where:
n is the number of calling attributes 25 being used to determine the tiestrengths
26; and
, 2, . . . , n are constants that may be derived, for example, from
historical data or the like.
At step 32, the social network analysis module 18, shown in FIG. 1, may then
calculate the tie-strengths 26 between subscribers 22 as a function w{x) of the
weighted sums x of the normalized attributes x between the subscribers 22. The
function w{x) is preferably a monotonically increasing function so that tiestrengths
26 between subscribers 22 are greater for greater weighted sums x .
For instance, in an exemplary embodiment, the social network analysis module
18, shown in FIG. 1, may determine the tie-strength 26 between two subscribers
22 using the monotonically increasing function w{x) given by:
w(x) =l -exp(-x/e2)
where:
w{x) is restricted to the interval [0,1 ] ; and
e is a constant parameter controlling the rate of saturation that may also be
derived, for example, from historical data or the like.
This exemplary function x ) is based on the assumption that once a strong
social connection is manifested between two subscribers 22, i.e. the tie-strength
26 is high, there is high probability that an idea, e.g. churning, will be transferred
from one subscriber 22 to the other subscriber 22. However, as should be
understood by those skilled in the art, this exemplary function x ) is merely
provided for illustrative purposes and those skilled in the art should readily
understand that a variety of other functions may be appropriate for correlating the
calling attributes 25 and determining the tie-strengths 26 between the subscribers
22.
Using the tie-strengths 26 and the churner data 24, shown in FIG. 1, the social
network analysis module 18, shown in FIG. 1, determines the churner influence
on the subscribers 22 at 34. The quantification of tie-strengths 26, as discussed
above, provides a basis for measuring which subscribers 22 are closely
connected to each other and, therefore, more likely to be influenced by each
other's behaviors.
For example, referring to FIG. 4 , the social network analysis module 18, shown in
FIG. 1, may use the tie-strengths 26 to model propagation of influence between
subscribers 22 to quantify how influence travels from a churner 36 to a social
circle 38 of the churner 36 and what portion of the influence is retained by
recipients 40 of the influence . To model the influence propagation through the
social circle, the social network analysis module 18, shown in FIG. 1, considers
the tie-strengths 26 between subscribers 22, denoted by nodes
etc. in the graph of FIG. 4 . Between a particular node n i and an adjoining node
ri j , the tie-strength 26 may be quantified as t For example, the tie-strength 26
between nodes nA and nB may be quantified as tAB and calculated as a function
of the calling attributes 25, shown in FIG. 1, between subscriber nodes nA and n
using the monotonically increasing function x ) discussed above. Since the
calling attributes 25, shown in FIG. 1, between two subscribers 22 are
undirectional between the subscribers 22, tie-strengths 26 are undirectional, i.e.
t = t
p
...
Once the tie-strengths 26 have been quantified, the social network analysis
module 18, shown in FIG. 1, may determine the influence l received by node n
from node n as a proportion of the tie-strength quantity t j between the nodes n
and ri j to the sum T of the tie-strength quanties of all ties incident on node n i
through the equation:
where:
I is the influence at node n ,
N is the set of all node neighbors of node n
For example, still referring to FIG. 4 , where node nA is a churner 36 having
influence , the influence received by node n from node nA may be quantified as
follows:
The total influence l i received by node n i is the sum of all of the influences
received from all of the neighbors of node n and given by the equation:
Thus, in keeping with the exemplary embodiment of FIG. 4 , the total influence at
node nB is I B = I BA + I BC since node nB will receive portions of the influence from
node nA in two ways, directly from node nA as I BA and indirectly through node nc ,
which receives a portion of the influence from node nA as I CA . The amount of
influence that node nB receives from nodes nA and nc is different because of the
relative tie-strengths 26 between each of the nodes. Each of the receiving
subscribers 22 will retain a portion of the influence received, pursuant to the
equations discussed above, and then pass that retained influence on to
neighboring subscribers 22. For example, subscriber node nE will receive a
portion of the influence of churn event of node nA via nodes nc and nD . This
propagation of influence / will continue until the retained influence reaches a
negligible quantity, e.g. the retained influence approximates zero, or until a
maximum pre-defined number of hops between subscribers 22 is reached. For
example, the maximum pre-defined number of hops may be set to 3 or 4 hops
between subscribers 22. In some embodiments, the social network analysis
module 18, shown in FIG. 1, may also include a parameter for decaying the
amount of influence propagated with the number of hops and/or with time.
Additionally, in some embodiments, the social network analysis module 18, shown
in FIG. 1, may consider a directionality of influence propagation with asymmetrical
tie-strengths 26.
The social network analysis module 18, shown in FIG. 1, may also limit the
propagation of influence / such that a particular subscriber 22 is not influenced by
the same churn event from the same neighboring subscriber 22 more than once.
Additionally, as should be understood by those skilled in the art, other limits
and/or constraints for propagating the influence / may be set through the social
network analysis module 18, shown in FIG. 1, when propagating the influence /
through the social circle 38 depending upon the desired model of the social ties
between subscribers 22.
The influence propagation calculation is repeated by the social network analysis
module 18, shown in FIG. 1, for every churn event that occurs for the network
operator 16, shown in FIG. 1. At the end of the propagation process, each of the
subscribers 22 will have a net amount of influence l gathered due to all of the
churn events in the social circle 38.
The exemplary embodiment described above provides a receiver-centric model
for influence propagation wherein the influence retained by the receiving
subscriber 22 is advantageously dependent upon the relationship between the
receiving subscriber 22 and the sending subscriber 22. For instance, if the
receiving subscriber 22 is a close friend of the sending subscriber 22, the tiestrength
26 between the receiving subscriber 22 and the sending subscriber 22
will be larger than the tie-strengths with some other neighbors of the receiving
subscriber 22, such as a colleague or an acquaintance. This larger tie-strength
26 will result in a larger influence at the receiving subscriber 22 from actions by
the sending subscriber 22 than from actions of the other neighbors, e.g. the
colleague or the acquaintance. The social network analysis module 18, shown in
FIG. 1, is adapted to account for these social differences by quantifying the tiestrengths
26 between subscribers 22 so that the total amount of retained influence
is relative to the tie-strength 26 that the receiving subscriber 22 has with the
sending subscriber 22 in relation to the tie-strengths 26 with all his neighboring
subscribers 22.
Referring back to FIG. 3, once the net amount of influence l i for each subscriber
22, shown in FIG. 4 , has been determined, the social network analysis module 18,
shown in FIG. 1, may provide socially relevant metrics for each subscriber 22,
shown in FIG. 4 , based on the social circle 38, shown in FIG. 4 , to the churn
prediction module 20 at step 42, as factors for determining the churn prediction
12. The socially relevant metrics for each subscriber 22, shown in FIG. 4 , may
include the net influence ; , the number of neighboring nodes n i that are churners
36, shown in FIG. 4 , the number of hops to the nearest churner 36, shown in FIG.
4 , the number/volume of calls to churners 36, shown in FIG. 4 , the
number/volume of calls to the nearest churners 36, shown in FIG. 4 , time spent on
calls to churners 36, shown in FIG. 4 , or other similar social data that may be
derived from the social circle 38, shown in FIG. 4 , and the call detail records 14,
shown in FIG. 1.
Additionally, in some embodiments, the social network analysis module 18, shown
in FIG. 1, may mine text from social media websites to glean information about
the sentiments that subscribers 22 have toward their network operator 16, shown
in FIG. 1. The social network analysis module 18, shown in FIG. 1, may use this
information linking each subscriber 22 to their social identity on the social
websites as a further social metric for generating the churn prediction 12.
At 44, the churn prediction module 20 generates the churn prediction 12 based on
the socially relevant metrics provided by the social network analysis module 18,
shown in FIG. 1, in addition to traditional subscriber-level metrics 46, shown in
FIG. 1. The traditional subscriber-level metrics 46, shown in FIG. 1, may include,
for example, service usage, billing, customer relationship management data (e.g.,
calls to customer support, outcome of complaints, demographic data and the like).
To generate the churn prediction 12, the churn prediction module 20 may
incorporate the socially relevant metrics provided by the social network analysis
module 18, shown in FIG. 1, and any traditional subscriber-level metrics 46,
shown in FIG. 1, as predictive variables in a traditional machine-learning algorithm
or process for predicting customer churn. For example, a number of different
classification algorithms and processes such as logistic regression, decision trees,
and random forests may be used for classifying the subscribers 22 as potential
churners. The classification process of the churn prediction module 20 may start
with a training data set, where the value of the target variable is known for the
subscribers 22, i.e., whether or not the each subscriber is a churner 36, shown in
FIG. 4 . The processes may use the training data to evaluate the relationship
between the predictive. Then, the churn prediction module 20 may use the
learned model to input and evaluate the predictive variables to generate the churn
prediction 12 indicating whether or not particular subscribers have a higher or
lower propensity to churn.
The churn prediction 12 may indicate a probability that each subscriber 22 will
churn, rather than just labeling each subscriber 22 as a potential churner or nonchurner,
so that subscribers 22 may be ordered from high to low churn likelihood
or propensity. With such ordering, the network operator 16 may advantageously
develop a retention campaign to target a limited number of subscribers 22 having
the highest likelihood or propensity to churn.
The computerized system 10, shown in FIG. 1, described herein has the
necessary electronics, software, memory, storage, databases, firmware,
logic/state machines, microprocessors, communication links, displays or other
visual or audio user interfaces, printing devices, and any other input/output
interfaces to perform the functions described herein and/or to achieve the results
described herein. For example, referring to FIG. 5, an exemplary embodiment of
the computerized system 10 is shown connected to the network operator 16
through a network interface unit 48. The computerized system 10 may include at
least one central processing unit (CPU) 50, system memory 52, including random
access memory (RAM) 54 and read-only memory (ROM) 56, an input/output
controller 58, and one or more data storage devices 60. All of these latter
elements are in communication with the CPU 50 to facilitate the operation of the
computerized system 10 as discussed above. Suitable computer program code
may be provided for executing numerous functions, including those discussed
above in connection with the social network analysis module 18 and churn
prediction module 20, both shown in FIG. 1. The computer program code may
also include program elements such as an operating system, a database
management system and "device drivers" that allow the CPU 50 to interface with
computer peripheral devices (e.g., a video display, a keyboard, a computer
mouse, etc.) via the input/output controller 58.
The CPU 50 may comprise a processor, such as one or more conventional
microprocessors and one or more supplementary co-processors such as math co
processors or the like. The CPU 50 is in communication with the network
interface unit 48, through which the CPU 50 may communicate with the network
operator 16 and/or other devices such as other servers or user terminals. The
network interface unit 48 may include multiple communication channels for
simultaneous communication with, for example, other processors, servers or
operators. Devices in communication with each other need not be continually
transmitting to each other. On the contrary, such devices need only transmit to
each other as necessary, may actually refrain from exchanging data most of the
time, and may require several steps to be performed to establish a communication
link between the devices.
[0001] The CPU 50 is in communication with the data storage device 60. The
data storage device 60 may comprise an appropriate combination of magnetic,
optical and/or semiconductor memory, and may include, for example, RAM, ROM,
flash drive, an optical disc such as a compact disc and/or a hard disk or drive.
The CPU 50 and the data storage device 60 each may be, for example, located
entirely within a single computer or other computing device; or connected to each
other by a communication medium, such as a USB port, serial port cable, a
coaxial cable, an Ethernet type cable, a telephone line, a radio frequency
transceiver or other similar wireless or wired medium or combination of the
foregoing. For example, the CPU 50 may be connected to the data storage
device 60 via the network interface unit 48.
The data storage device 60 may store, for example, one or more databases
adapted to store information that may be utilized to store information required by
the program, an operating system for the computerized system 10, and/or one or
more programs (e.g., computer program code and/or a computer program
product) adapted to direct the CPU 50 to generate churn predictions 12, shown in
FIG. 1. The operating system and/or programs may be stored, for example, in a
compressed, an uncompiled and/or an encrypted format, and may include
computer program code. The instructions of the computer program code may be
read into a main memory of the processor from a computer-readable medium
other than the data storage device 60, such as from the ROM 56 or from the RAM
54. While execution of sequences of instructions in the program causes the
processor to perform the process steps described herein, hard-wired circuitry may
be used in place of, or in combination with, software instructions for
implementation of the processes of the present invention. Thus, embodiments of
the present invention are not limited to any specific combination of hardware and
software.
The program may also be implemented in programmable hardware devices such
as field programmable gate arrays, programmable array logic, programmable
logic devices or the like. Programs may also be implemented in software for
execution by various types of computer processors. A program of executable
code may, for instance, comprise one or more physical or logical blocks of
computer instructions, which may, for instance, be organized as an object,
procedure, process or function. Nevertheless, the executables of an identified
program need not be physically located together, but may comprise separate
instructions stored in different locations which, when joined logically together,
comprise the program and achieve the stated purpose for the programs such as
generating churn predictions 12, shown in FIG. 1. In an embodiment, an
application of executable code may be a compilation of many instructions, and
may even be distributed over several different code partitions or segments,
among different programs, and across several devices.
The term "computer-readable medium" as used herein refers to any medium that
provides or participates in providing instructions to the CPU 50 of the
computerized system 10 (or any other processor of a device described herein) for
execution. Such a medium may take many forms, including but not limited to,
non-volatile media and volatile media. Non-volatile media include, for example,
optical, magnetic, or opto-magnetic disks, such as memory. Volatile media
include dynamic random access memory (DRAM), which typically constitutes the
main memory. Common forms of computer-readable media include, for example,
a floppy disk, a flexible disk, hard disk, magnetic tape, any other magnetic
medium, a CD-ROM, DVD, any other optical medium, punch cards, paper tape,
any other physical medium with patterns of holes, a RAM, a PROM, an EPROM
or EEPROM (electronically erasable programmable read-only memory), a FLASHEEPROM,
any other memory chip or cartridge, or any other medium from which a
computer can read.
Various forms of computer readable media may be involved in carrying one or
more sequences of one or more instructions to the CPU 50 (or any other
processor of a device described herein) for execution. For example, the
instructions may initially be borne on a magnetic disk of a remote computer (not
shown). The remote computer can load the instructions into its dynamic memory
and send the instructions over an Ethernet connection, cable line, or even
telephone line using a modem. A communications device local to a computing
device (e.g., a server) can receive the data on the respective communications line
and place the data on a system bus for the CPU 50. The system bus carries the
data to main memory, from which the CPU 50 retrieves and executes the
instructions. The instructions received by main memory may optionally be stored
in memory either before or after execution by the CPU 50. In addition,
instructions may be received via a communication port as electrical,
electromagnetic or optical signals, which are exemplary forms of wireless
communications or data streams that carry various types of information.
The computerized system 10 advantageously provides a system and method to
predict customer churn in telecommunication services that integrates social
network analysis concepts with traditional churn prediction systems and methods
in an effort to detect potential churners before they unsubscribe from network
operators. The tie-strengths 26, shown in FIG. 4 , and the influence propagation
model developed by the social network analysis module 18, shown in FIG. 1, may
beneficially be integrated into the machine-learning based churn prediction
module 20, shown in FIG. 1, to improve churn prediction accuracy. Once
potential churners are identified, they may then be targeted with retention
campaigns and the like.
Additionally, the system and method of the computerized system 10 may
advantageously be applied to model influence diffusion for a variety of social
phenomenon that may influence a subscriber 22, shown in FIG. 1, such as upselling
of services, cross-selling of services and downloading of applications. For
example, the social network analysis module 18, shown in FIG. 1, may be
implemented to better target current subscribers 22, shown in FIG. 1, for new
services and/or applications.
Although this invention has been shown and described with respect to the detailed
embodiments thereof, it will be understood by those skilled in the art that various
changes in form and detail thereof may be made without departing from the spirit
and the scope of the invention.

What is claimed is:
1. A system for generating a subscriber churn prediction comprising:
a data input device adapted to receive call detail records from a network
operator detailing communication for at least one subscriber of the network
operator; and
at least one processor connected to the data input device, the at least one
processor adapted to execute a churn prediction program to generate the
subscriber churn prediction based at least on the call detail records.
2. The system according to claim 1, wherein the churn prediction program
includes a social network analysis module that derives social metrics from the call
detail records for generating the churn prediction.
3. The system according to claim 2, wherein the churn prediction program
includes a churn prediction module that receives the social metrics from the social
network analysis module and generates the churn prediction.
4 . The system according to claim 3, wherein the churn prediction is generated
from the social metrics in combination with at least one of a service usage metric,
a billing metric, a number of calls to customer support, an outcome of complaints
or a demographic data.
5. The system according to claim 2, wherein the social network analysis module
derives at least one social metric by determining tie-strengths between connected
subscribers of the network operator based on one or more calling attributes of the
call detail records.
6. The system according to claim 5, wherein the tie-strengths are determined
based on an average of a plurality of calling attributes of the call detail records.
7. The system according to claim 5, wherein the at least one social metric
includes at least one of a net churner influence, a number of neighboring
subscribers that are churners, a number of hops to a nearest churner, a number
of calls to churners, a number of calls to the nearest churner, and a time spent on
calls to churners.
8. The system according to claim 5, wherein the social network analysis module
propagates a churner influence between subscribers based on the tie-strength
and determines a net influence for each subscriber.
9. The system according to claim 8, wherein the propagation of the churner
influence is receiver centric with the influence received by each subscriber from a
connected subscriber being proportional to the tie-strength of the tie between the
subscribers.
10. The system according to claim 8, wherein the social network analysis module
limits the propagation of the churner influence based on a number of connections
between subscribers.