Machine learning-based cellular service issue troubleshooting using limited ground truth data
Aspects of the subject disclosure may include, for example, training a first machine learning model based on a combination of labeled training data and unlabeled training data, the first machine learning model producing augmented training data, training a second machine learning model based on a combination of the labeled training data and the augmented training data, receiving, at a client device, customer information about a service degradation at a user equipment (UE) device of a customer in a cellular network, providing the customer information to the second machine learning model, receiving, at the client device, from the second machine learning model, information identifying a root cause of the service degradation, and modifying a network component of the cellular network or the UE device, based on the information identifying a root cause of the service degradation. Other embodiments are disclosed.
1 . A device, comprising:
a processing system including a processor; and
a memory that stores executable instructions that, when executed by the processing system, facilitate performance of operations, the operations comprising:
training a first machine learning model based on a combination of labeled training data and unlabeled training data, the first machine learning model producing augmented training data;
identifying future features, wherein each respective future feature of the future features corresponds to a resolution action taken in an offline phase to resolve a reported service degradation after an online phase corresponding to a past customer interaction, wherein identifying future features comprises:
comparing a previous network condition before a particular care call of a plurality of previous care calls with a subsequent network condition after the particular care call,
in response to determining that UE-level performance improved substantially immediately after the particular care call, inferring that a root cause of the reported service degradation was a user device, and
in response to determining that UE-level performance did not improve substantially immediately after the particular care call, inferring that the root cause of the reported service degradation was a network problem;
adding the future features to the augmented training data;
training a second machine learning model based on a combination of the labeled training data and the augmented training data;
receiving customer information about a current service degradation at a user equipment (UE) device of a customer in a cellular network;
providing the customer information to the second machine learning model;
receiving from the second machine learning model, information identifying a root cause of the current service degradation; and
modifying a network component of the cellular network or the UE device, based on the information identifying a root cause of the current service degradation.
2 . The device of claim 1 , wherein identifying future features further comprises:
retrieving information about the plurality of previous care calls from customers,
wherein the particular care call reports the reported service degradation at a customer UE device, and wherein the subsequent network condition occurs after troubleshooting actions are taken based on the particular care call; and
inferring whether the troubleshooting actions corrected the reported service degradation.
3 . The device of claim 2 , wherein the operations further comprise:
identifying information about the particular care call in the unlabeled training data.
4 . The device of claim 2 , wherein identifying future features further comprises:
determining that a UE-performance improvement correlates with a change in a particular network component of the cellular network, wherein
inferring that the root cause of the reported service degradation was a network problem is responsive to the determining that the UE-performance improvement correlates with the change in a particular network component.
5 . The device of claim 2 , wherein the operations further comprise:
inferring, based on the comparing the previous network condition with the subsequent network condition, a beginning time and an ending time for the reported service degradation.
6 . The device of claim 2 , wherein the operations further comprise:
inferring, based on the comparing the previous network condition with the subsequent network condition, whether performance of the customer UE device improves after the particular care call.
7 . The device of claim 2 , wherein the comparing the previous network condition with the subsequent network condition comprises:
comparing key performance indicators for a time period before the particular care call with the key performance indicators for a time period after the particular care call.
8 . The device of claim 1 , wherein the identifying future features comprises:
retrieving historical data for the cellular network;
identifying, in the historical data for the cellular network, one or more key performance indicators in the cellular network prior to the reported service degradation;
identifying, in the historical data for the cellular network, an improvement in the one or more key performance indicators during a time period subsequent to the reported service degradation; and
adding information about the one or more key performance indicators to the future features.
9 . The device of claim 1 , wherein the first machine learning model comprises a teacher model used exclusively for facilitating the training the second machine learning model by providing the augmented training data for the second machine learning model in response to the adding the future features to the augmented training data, and wherein the second machine learning model comprises a student model.
10 . The device of claim 1 , wherein modifying the network component of the cellular network comprises repairing or replacing the network component.
11 . A non-transitory, machine-readable medium, comprising executable instructions that, when executed by a processing system including a processor, facilitate performance of operations, the operations comprising:
training a first machine learning model based on a combination of labeled training data and unlabeled training data, the first machine learning model producing augmented training data;
retrieving information about a plurality of previous care calls from customers;
comparing a previous network condition before a particular care call of the plurality of previous care calls with a subsequent network condition after the particular care call, the particular care call reporting a service degradation at a customer UE device, resulting in a reported service degradation, the subsequent network condition occurring after troubleshooting actions are taken based on the particular care call;
in response to determining that UE-level performance improved substantially immediately after the troubleshooting actions, inferring that a root cause of the reported service degradation associated with the particular care call was a user device;
in response to determining that UE-level performance did not improve substantially immediately after the troubleshooting actions, inferring that the root cause of the reported service degradation associated with the particular care call was a network problem;
training a second machine learning model based on a combination of the labeled training data, the augmented training data, and the inferring of the root cause;
receiving customer information about a current service degradation at a user equipment (UE) device of a customer in a cellular network;
providing the customer information to the second machine learning model;
receiving from the second machine learning model, information identifying a root cause of the current service degradation; and
modifying a network component of the cellular network or the UE device, based on the information identifying a root cause of the current service degradation.
12 . The non-transitory, machine-readable medium of claim 11 , wherein the operations further comprise:
determining that a UE-performance improvement correlates with a change in a particular network component of the cellular network, wherein
inferring that the root cause of the reported service degradation was a network problem is responsive to the determining that the UE-performance improvement correlates with the change in a particular network component.
13 . The non-transitory, machine-readable medium of claim 11 , wherein the first machine learning model comprises a teacher model, and the second machine learning model comprises a student model.
14 . The non-transitory, machine-readable medium of claim 11 , wherein modifying the network component of the cellular network comprises repairing or replacing the network component.
15 . A method, comprising:
training, by a processing system including a processor, a first machine learning model based on a combination of labeled training data and unlabeled training data, the first machine learning model producing augmented training data;
identifying, by the processing system, future features, wherein each respective future feature of the future features corresponds to a resolution action taken in an offline phase to resolve a reported service degradation after an online phase corresponding to a past customer interaction, wherein identifying future features comprises:
comparing a previous network condition before receiving the customer information with a subsequent network condition after receiving the customer information,
in response to determining that UE-level performance improved substantially immediately after receiving the customer information, inferring that the root cause of the reported service degradation was a user device, and
in response to determining that UE-level performance did not improve substantially immediately after receiving the customer information, inferring that the root cause of the reported service degradation was a network problem;
adding, by the processing system, the future features to the augmented training data;
training, by the processing system, a second machine learning model based on a combination of the labeled training data and the augmented training data;
receiving, by the processing system, customer information about a current service degradation at a user equipment (UE) device of a customer in a cellular network;
providing, by the processing system, the customer information to the second machine learning model;
receiving, by the processing system, from the second machine learning model, information identifying a root cause of the current service degradation; and
modifying, by the processing system, a network component of the cellular network or the UE device, based on the information identifying the root cause of the current service degradation.
16 . The method of claim 15 , wherein identifying future features further comprises:
retrieving, by the processing system, information about the plurality of previous care calls from customers,
wherein the particular care call reports the reported service degradation at a customer UE device, and wherein the subsequent network condition occurs after troubleshooting actions are taken based on the particular care call; and
inferring, by the processing system, whether the troubleshooting actions corrected the reported service degradation.
17 . The method of claim 16 , further comprising:
identifying, by the processing system, information about the particular care call in the unlabeled training data.
18 . The method of claim 16 , wherein identifying future features further comprises:
determining, by the processing system, that a UE-performance improvement correlates with a change in a particular network component of the cellular network, wherein
inferring, by the processing system, that the root cause of the reported service degradation was a network problem is responsive to the determining that the UE-performance improvement correlates with the change in a particular network component.
19 . The method of claim 15 , wherein the first machine learning model comprises a teacher model used exclusively for facilitating the training the second machine learning model by providing the augmented training data for the second machine learning model in response to the adding the future features to the augmented training data, and wherein the second machine learning model comprises a student model.
20 . The method of claim 15 , wherein modifying the network component of the cellular network comprises repairing or replacing the network component.