Automated request processing using ensemble machine learning framework

1 . A computer-implemented method comprising:

aggregating interaction data associated with a request;

generating, by a first machine learning model, predicted user sentiment related to the request based at least in part on a portion of the aggregated interaction data;

generating a first score for the request based at least in part on a comparison of the aggregated interaction data to a set of keywords, wherein the set of keywords is generated by intersecting words identified based at least in part on co-occurrence frequencies of words within historical aggregated interaction data corresponding to a plurality of historical requests with a manually curated list of keywords;

computing a second score for the request, wherein the second score is based on a weighted combination of at least the first score and the predicted user sentiment output from the first machine learning model;

determining, by a second machine learning model, whether the request is anomalous, wherein the second machine learning model obtains the second score as an input and determines whether the request is anomalous based at least in part on the second score; and

in response to determining that the request is anomalous:

(i) initiating one or more automated actions for the request; and

(ii) dynamically updating the set of keywords by extracting one or more new keywords from the aggregated interaction data of the anomalous request, wherein the one or more new keywords are identified based at least in part on co-occurrence frequencies of words within the aggregated interaction data of the anomalous request; and

determining whether an additional request is anomalous based at least in part on the updated set of keywords;

wherein the method is performed by at least one processing device comprising a processor coupled to a memory.

2 . The computer-implemented method of claim 1 , comprising:

obtaining order information for one or more of a service and a product associated with the request, wherein the computing the second score is further based on one or more features extracted from the order information, and wherein the one or more features comprise a change in delivery time, a part shortage, an increase in manufacturing lead time, and/or a supply chain disruption.

3 . The computer-implemented method of claim 1 , wherein the first machine learning model comprises a transformer-based machine learning model that generates scores for a plurality of emotions, and wherein the predicted user sentiment is determined by aggregating the scores for a designated subset of the plurality of emotions.

4 . The computer-implemented method of claim 1 , wherein the second machine learning model comprises a boosting machine learning process.

5 . The computer-implemented method of claim 1 , wherein the aggregated interaction data comprises at least one of: text data and voice data.

6 . The computer-implemented method of claim 1 , wherein the one or more automated actions comprises at least one of:

generating one or more notifications related to the anomalous request;

escalating the anomalous request for handling by one or more support agents; and

automatically adjusting an order associated with the request.

7 . The computer-implemented method of claim 1 , wherein the request is from a user and associated with one or more of a service and a product.

8 . A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device:

to aggregate interaction data associated with a request;

to generate, by a first machine learning model, predicted user sentiment related to the request based at least in part on a portion of the aggregated interaction data;

to generate a first score for the request based at least in part on a comparison of the aggregated interaction data to a set of keywords, wherein the set of keywords is generated by intersecting words identified based at least in part on co-occurrence frequencies of words within historical aggregated interaction data corresponding to a plurality of historical requests with a manually curated list of keywords;

to compute a second score for the request, wherein the second score is based on a weighted combination of at least the first score and the predicted user sentiment output from the first machine learning model;

to determine, by a second machine learning model, whether the request is anomalous, wherein the second machine learning model obtains the second score as an input and determines whether the request is anomalous based at least in part on the second score;

in response to determining that the request is anomalous:

(i) to initiate one or more automated actions for the request; and

(ii) to dynamically update the set of keywords by extracting one or more new keywords from the aggregated interaction data of the anomalous request, wherein the one or more new keywords are identified based at least in part on co-occurrence frequencies of words within the aggregated interaction data of the anomalous request; and

to determine whether an additional request is anomalous based at least in part on the updated set of keywords.

9 . The non-transitory processor-readable storage medium of claim 8 , wherein the program code when executed by the at least one processing device causes the at least one processing device:

to obtain order information for one or more of a service and a product associated with the request, wherein the computing the second score is further based on one or more features extracted from the order information, and wherein the one or more features comprise a change in delivery time, a part shortage, an increase in manufacturing lead time, and/or a supply chain disruption.

10 . The non-transitory processor-readable storage medium of claim 8 , wherein the first machine learning model comprises a transformer-based machine learning model that generates scores for a plurality of emotions, and wherein the predicted user sentiment is determined by aggregating the scores for a designated subset of the plurality of emotions.

11 . The non-transitory processor-readable storage medium of claim 8 , wherein the second machine learning model comprises a boosting machine learning process.

12 . The non-transitory processor-readable storage medium of claim 8 , wherein the aggregated interaction data comprises at least one of: text data and voice data.

13 . The non-transitory processor-readable storage medium of claim 8 , wherein the request is from a user and associated with one or more of a service and a product.

14 . An apparatus comprising:

at least one processing device comprising a processor coupled to a memory;

the at least one processing device being configured:

to aggregate interaction data associated with a request;

to generate, by a first machine learning model, predicted user sentiment related to the request based at least in part on a portion of the aggregated interaction data;

to generate a first score for the request based at least in part on a comparison of the aggregated interaction data to a set of keywords, wherein the set of keywords is generated by intersecting words identified based at least in part on co-occurrence frequencies of words within historical aggregated interaction data corresponding to a plurality of historical requests with a manually curated list of keywords;

to compute a second score for the request, wherein the second score is based on a weighted combination of at least the first score and the predicted user sentiment output from the first machine learning model;

to determine, by a second machine learning model, whether the request is anomalous, wherein the second machine learning model obtains the second score as an input and determines whether the request is anomalous based at least in part on the second score;

in response to determining that the request is anomalous:

(i) to initiate one or more automated actions for the request; and

(ii) to dynamically update the set of keywords by extracting one or more new keywords from the aggregated interaction data of the anomalous request, wherein the one or more new keywords are identified based at least in part on co-occurrence frequencies of words within the aggregated interaction data of the anomalous request; and

to determine whether an additional request is anomalous based at least in part on the updated set of keywords.

15 . The apparatus of claim 14 , wherein the at least one processing device is further configured:

to obtain order information for one or more of a service and a product associated with the request, wherein the computing the second score is further based on one or more features extracted from the order information, and wherein the one or more features comprise a change in delivery time, a part shortage, an increase in manufacturing lead time, and/or a supply chain disruption.

16 . The apparatus of claim 14 , wherein the first machine learning model comprises a transformer-based machine learning model that generates scores for a plurality of emotions, and wherein the predicted user sentiment is determined by aggregating the scores for a designated subset of the plurality of emotions.

17 . The apparatus of claim 14 , wherein the second machine learning model comprises a boosting machine learning process.

18 . The apparatus of claim 14 , wherein the aggregated interaction data comprises at least one of: text data and voice data.

19 . The apparatus of claim 14 , wherein the one or more automated actions comprises at least one of:

generating one or more notifications related to the anomalous request;

escalating the anomalous request for handling by one or more support agents; and

automatically adjusting an order associated with the request.

20 . The apparatus of claim 14 , wherein the request is from a user and associated with one or more of a service and a product.