Real-time user communication sentiment detection for dynamic anomaly detection and mitigation

1 . At least one non-transitory, computer-readable storage medium storing instructions, which, when executed by at least one data processor of a system, cause the system to:

obtain a textual representation of an interaction between a user of a wireless telecommunication network and an operator of the wireless telecommunication network;

obtain a criterion comprising multiple keywords indicating positive user sentiment;

input the textual representation of the interaction and the obtained criterion to an artificial intelligence model configured to analyze natural language to determine, by the artificial intelligence model, a timeseries indicating a set of user sentiment values during the interaction at different points in time,

wherein each user sentiment value of the set of user sentiment values represents a satisfied sentiment or an unsatisfied sentiment using a numerical range, and

wherein determining the timeseries includes:

separating the textual representation into multiple portions wherein a particular portion among the multiple portions occurs within a continuous time period;

determining a proportion value characterizing a proportion of words of the textual representation that are associated with the multiple keywords indicating the positive user sentiment;

determining, using the proportion value and by the artificial intelligence model, a particular user sentiment value associated with the particular portion among the multiple portions; and

creating the timeseries correlating the continuous time period and the set of user sentiment values;

determine whether the timeseries includes an anomaly by determining whether at least a portion of the timeseries satisfies the criterion; and

upon determining that the timeseries includes the anomaly, transmit a notification indicating that the operator of the wireless telecommunication network generate a recommendation;

determine one or more available assisting operators comprising at least one machine learning model associated with at least one controller;

in response to determining the one or more available assisting operators, determine, using an access and mobility management function (AMF) component of the wireless telecommunication network and by querying a network function repository function (NRF), an indication of one or more communication links associated with the one or more available assisting operators;

in response to determining the indication of the one or more communication links, dynamically configuring, using a session management function (SMF) component, the one or more communication links between the one or more available assisting operators and the user to enable assistance of the user,

wherein the one or more available assisting operators include the at least one machine learning model;

based on determining that the timeseries includes the anomaly, generate a report for the interaction using the timeseries;

based on an identifier of the operator, determine the at least one controller including a training algorithm; and

transmit the report to the determined at least one controller to enable training of the one or more available assisting operators using the training algorithm of the at least one controller.

2 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

obtain the criterion including obtaining a first threshold indicating an amount of time and a second threshold indicating a number within the numerical range;

determine whether the timeseries includes a period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied; and

upon determining that the timeseries includes the period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied, determine that the timeseries includes the anomaly.

3 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

obtain the criterion including a set of keywords indicating that the user is unsatisfied;

determine whether the textual representation of the interaction includes a keyword among the set of keywords; and

upon determining that the textual representation of the interaction includes the keyword among the set of keywords, determine that the timeseries includes the anomaly.

4 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

obtain the criterion including a first threshold indicating a change in the user sentiment values, and an amount of time;

determine whether the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time; and

upon determining that the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time, determine that the timeseries includes the anomaly.

5 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

determine an average sentiment of the timeseries,

wherein the average sentiment of the timeseries represents an arithmetic mean of sentiments associated with the multiple portions of the textual representation of the interaction;

compare the average sentiment of the timeseries with a plurality of average sentiments associated with a plurality of operators of the wireless telecommunication network; and

based on comparing the average sentiment of the timeseries with each average sentiment of the plurality of average sentiments, generate a ranking of the operator with respect to the plurality of operators.

6 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

based on the determining that the timeseries includes the anomaly, generate a report for the interaction using the timeseries,

wherein the report comprises one or more of:

an indication of an average sentiment of the timeseries,

an indication of the anomaly, and

one or more keywords included in the textual representation indicating that the user is unsatisfied;

based on an identifier of the operator, determine an associated controller; and

transmit the report to the associated controller to enable the associated controller to train the operator.

7 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

based on determining that the timeseries includes the anomaly, determine a plurality of assisting operators associated with the operator;

transmit the notification indicating that the operator needs assistance to the plurality of assisting operators; and

based on receiving, from one or more assisting operators of the plurality of assisting operators, one or more responses indicating availability, configure communication links between the one or more assisting operators and the user to enable assistance of the user.

8 . The at least one non-transitory, computer-readable medium of claim 1 , including instructions to:

determine a plurality of timeseries corresponding to a plurality of interactions between users of the wireless telecommunication network and operators of the wireless telecommunication network,

wherein timeseries of the plurality of timeseries represent sentiments associated with users during corresponding interactions at different points in time;

determine a plurality of average sentiments corresponding to the plurality of timeseries; and

based on the plurality of average sentiments, determine a network satisfaction metric and a predicted attrition rate,

wherein the network satisfaction metric indicates an average sentiment across the plurality of interactions, and

wherein the predicted attrition rate indicates a predicted percentage of the users that will disengage with the wireless telecommunication network.

9 . A system comprising:

at least one hardware processor; and

at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:

obtain an indication of an interaction between a user of a wireless telecommunication network and an operator of the wireless telecommunication network;

obtain a criterion comprising multiple keywords indicating positive user sentiment;

determine a timeseries indicating a set of user sentiment values associated with the user during the interaction at different points in time,

wherein each user sentiment value of the set of user sentiment values represents a satisfied sentiment or an unsatisfied sentiment using a numerical range, and

wherein determining the timeseries includes:

separating the indication of the interaction into multiple portions wherein a particular portion among the multiple portions occurs within a continuous time period;

determining a proportion value characterizing a proportion of words of the indication of the interaction that are associated with the multiple keywords indicating the positive user sentiment;

determining, using the proportion value, a particular user sentiment value associated with the particular portion among the multiple portions; and

creating the timeseries correlating the continuous time period and the set of user sentiment values;

determine whether the timeseries includes an anomaly by determining whether at least a portion of the timeseries satisfies the criterion;

upon determining that the timeseries includes the anomaly, create a notification indicating that the operator of the wireless telecommunication network generate a recommendation;

determine one or more available assisting operators comprising at least one machine learning model associated with at least one controller;

in response to determining the one or more available assisting operators, determine, using an AMF component of the wireless telecommunication network and by querying an NRF, an indication of one or more communication links associated with the one or more available assisting operators;

in response to determining the indication of the one or more communication links, dynamically configuring, using an SMF component, the one or more communication links between the one or more available assisting operators and the user to enable assistance of the user,

wherein the one or more available assisting operators include the at least one machine learning model;

based on determining that the timeseries includes the anomaly, generate a report for the interaction using the timeseries;

based on an identifier of the operator, determine the at least one controller including a training algorithm; and

transmit the report to the determined at least one controller to enable training of the one or more available assisting operators using the training algorithm of the at least one controller.

10 . The system of claim 9 , wherein instructions cause the system to:

obtain the criterion including obtaining a first threshold indicating an amount of time and a second threshold indicating a number within the numerical range;

determine whether the timeseries includes a period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied; and

upon determining that the timeseries includes the period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied, determine that the timeseries includes the anomaly.

11 . The system of claim 9 , wherein the instructions cause the system to:

obtain the criterion including a set of keywords indicating that the user is unsatisfied;

determine whether the indication of the interaction includes a keyword among the set of keywords; and

upon determining that the indication of the interaction includes the keyword among the set of keywords, determine that the timeseries includes the anomaly.

12 . The system of claim 9 , wherein the instructions cause the system to:

obtain the criterion including a first threshold indicating a change in the user sentiment values, and an amount of time;

determine whether the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time; and

upon determining that the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time, determine that the timeseries includes the anomaly.

13 . The system of claim 9 , wherein the instructions cause the system to:

determine an average sentiment of the timeseries,

wherein the average sentiment of the timeseries represents an arithmetic mean of sentiments associated with the multiple portions of the indication of the interaction;

compare the average sentiment of the timeseries with a plurality of average sentiments associated with a plurality of operators of the wireless telecommunication network; and

based on comparing the average sentiment of the timeseries with each average sentiment of the plurality of average sentiments, generating a ranking of the operator with respect to the plurality of operators.

14 . The system of claim 9 , wherein the instructions cause the system to:

based on determining that the timeseries includes the anomaly, generate a report for the interaction using the timeseries,

wherein the report comprises one or more of:

an indication of an average sentiment of the timeseries,

an indication of the anomaly, and

one or more keywords included in the indication of the interaction indicating that the user is unsatisfied;

based on an identifier of the operator, determine an associated controller; and

transmit the report to the associated controller to enable the associated controller to train the operator.

15 . The system of claim 9 , wherein the instructions cause the system to:

based on determining that the timeseries includes the anomaly, determine a plurality of assisting operators associated with the operator;

transmit the notification indicating that the operator of the wireless telecommunication network needs assistance to the plurality of assisting operators; and

based on receiving, from one or more assisting operators of the plurality of assisting operators, one or more responses indicating availability, configure communication links between the one or more assisting operators and the user to enable assistance of the user.

16 . The system of claim 9 , wherein the instructions cause the system to:

determine a plurality of timeseries corresponding to a plurality of interactions between users of the wireless telecommunication network and operators of the wireless telecommunication network,

wherein timeseries of the plurality of timeseries represent sentiments associated with users during corresponding interactions at different points in time;

determine a plurality of average sentiments corresponding to the plurality of timeseries; and

based on the plurality of average sentiments, determine a network satisfaction metric and a predicted attrition rate,

wherein the network satisfaction metric indicates an average sentiment across the plurality of interactions, and

wherein the predicted attrition rate indicates a predicted percentage of the users that will disengage with the wireless telecommunication network.

17 . A method comprising:

obtaining an indication of an interaction between a user of a wireless telecommunication network and an operator of the wireless telecommunication network;

obtain a criterion comprising multiple keywords indicating positive user sentiment;

determining a timeseries indicating a set of user sentiment values associated with the user during the interaction at different points in time,

wherein the user sentiment value of the set of user sentiment values represents a satisfied sentiment or an unsatisfied sentiment using a numerical range, and

wherein determining the timeseries includes:

separating the indication of the interaction into multiple portions wherein a portion among the multiple portions occurs within a continuous time period;

determining a proportion value characterizing a proportion of words of the indication of the interaction that are associated with the multiple keywords indicating the positive user sentiment;

determining, using the proportion value, a particular user sentiment value associated with the particular portion among the multiple portions; and

creating the timeseries correlating the continuous time period and the set of user sentiment values;

determining whether the timeseries includes an anomaly by determining whether at least a portion of the timeseries satisfies the criterion;

upon determining that the timeseries includes the anomaly, creating a notification indicating that the operator of the wireless telecommunication network generate a recommendation;

determining one or more available assisting operators comprising at least one machine learning model associated with at least one controller;

in response to determining the one or more available assisting operators, determine, using an AMF component of the wireless telecommunication network and by querying an NRF, an indication of one or more communication links between the one or more available assisting operators and the user to enable assistance of the user,

wherein the one or more available assisting operators include the at least one machine learning model;

based on determining that the timeseries includes the anomaly, generate a report for the interaction using the timeseries;

based on an identifier of the operator, determine the at least one controller including a training algorithm; and

transmit the report to the determined at least one controller to enable training of the one or more available assisting operators using the training algorithm of the at least one controller.

18 . The method of claim 17 , comprising:

obtaining the criterion including obtaining a first threshold indicating an amount of time and a second threshold indicating a number within the numerical range;

determining whether the timeseries includes a period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied; and

upon determining that the timeseries includes the period of time equal to or greater than the first threshold during which the set of user sentiment values and the second threshold indicate that the user is unsatisfied, determining that the timeseries includes the anomaly.

19 . The method of claim 17 , comprising:

obtaining the criterion including a set of keywords indicating that the user is unsatisfied;

determining whether the indication of the interaction includes a keyword among the set of keywords; and

upon determining that the indication of the interaction includes the keyword among the set of keywords, determining that the timeseries includes the anomaly.

20 . The method of claim 17 , comprising:

obtaining the criterion including a first threshold indicating a change in the user sentiment values, and an amount of time;

determining whether the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time; and

upon determining that the timeseries includes the first threshold wherein the change in the user sentiment values lasts at least the amount of time, determining that the timeseries includes the anomaly.