Machine-learning predictive models for classifying responses to and outcomes of end-user communications
A service provider computing system includes memory having stored thereon instructions that, when executed by one or more processors, cause the one or more processors to obtain a first set of complaint logs corresponding to one or more user complaints, each complaint log including a resolution comments field providing a textual representation of a resolution for a corresponding complaint log, parse the resolution comments field of each complaint log of the first set of complaint logs to identify one or more key terms, and execute a machine-learning predictive model using the one or more key terms to generate, for each complaint log, a prediction indicating whether a corresponding complaint should have compensation, the machine-learning predictive model being a supervised machine learning model configured to accept the parsed resolution comments field for the first set of complaint logs as inputs and to output the prediction for each complaint log.
1. A service provider computing system comprising:
memory having stored thereon instructions that, when executed by one or more processors, cause the one or more processors to:
obtain a first set of complaint logs corresponding to one or more user complaints, each complaint log of the first set of complaint logs comprising a resolution comments field and an indication whether compensation was provided, the resolution comments field providing a textual representation of a resolution for a corresponding complaint log;
parse the resolution comments field of each complaint log of the first set of complaint logs to identify one or more key terms; and
execute a machine-learning predictive model using the one or more key terms to generate, for each complaint log, a prediction indicating whether a corresponding complaint should have compensation, the machine-learning predictive model being a supervised machine learning model configured to accept the parsed resolution comments field for the first set of complaint logs as inputs and to output the prediction for each complaint log.
2. The system of claim 1 , wherein parsing the resolution comments comprises generating a unigram matrix, the unigram matrix indicating a frequency of each unigram.
3. The system of claim 1 , wherein the instructions further cause the one or more processors to generate a report for the first set of complaint logs, the report including a probability that each complaint associated with the first set of complaint logs should have compensation.
4. The system of claim 3 , wherein the report includes only complaint logs where the probability exceeds a threshold range.
5. The system of claim 3 , wherein report further indicates whether compensation was actually provided for each complaint.
6. The system of claim 1 , wherein parsing the resolution comments field comprises at least one of removing any numerical values, removing any repeated words, or identifying a root of each word.
7. The system of claim 1 , wherein executing the machine-learning predictive model comprises generating, for each complaint log, a probability that the user should receive compensation.
8. The system of claim 1 , wherein the instructions further cause the one or more processors to:
compare the prediction for each complaint log to the indication whether compensation was provided; and
in response to determining, for at least one complaint log, that the prediction does not agree with the indication, identifying a plurality of similar complaints to the at least one complaint log.
9. The system of claim 8 , wherein at least five additional complaint logs are identified.
10. The system of claim 1 , wherein the instructions further cause the one or more processors to identify, for each complaint log, at least one additional complaint log that is similar to a corresponding complaint log based on the one or more key terms.
11. The system of claim 1 , wherein the instructions further cause the one or more processors to:
train the machine-learning predictive model using the first set of complaint logs;
obtain a second set of complaint logs, each complaint log of the second set of complaint log comprising a resolution comments field and an indication whether compensation was provided;
parse, for each complaint log of the second set of complaint logs, the resolution comments field to generate a unigram matrix for the second set of complaint logs; and
execute the trained machine-learning predictive model based on the unigram matrix for the second set of complaint logs to generate a prediction indicating whether compensation should be provided for each complaint log of the second set of complaint logs.
12. The system of claim 1 , wherein the machine-learning predictive model comprises a sequential neural network.
13. The system of claim 1 , wherein the machine-learning predictive model comprises a bidirectional encoder representations from transformers (BERT) model.
14. The system of claim 1 , wherein the instructions further cause the one or more processors to present, via a graphical user interface, the prediction for each of the complaint logs.
15. A method comprising:
receiving a complaint from a user;
recording at least a text-based description of the complaint, including an indication of a resolution for the complaint;
parsing the text-based description of the complaint to generate a matrix of key terms within the text-based description;
executing a machine-learning predictive model using the matrix of key terms to generate, for the complaint, a prediction indicating whether the complaint should have compensation;
presenting, via a user interface, an indication of the prediction;
determining, based on the parsed and text-based description of the complaint, an indication regarding whether compensation was actually provided;
comparing the prediction to the indication; and
in response to determining that the prediction does not agree with the indication, identifying a plurality of additional complaints that are similar to the first complaint based on a text-based description of each of the additional complaints.
16. The method of claim 15 , wherein the machine-learning predictive model is a supervised machine learning model configured to accept the parsed text-based description as an input and to output the prediction as a probability that the complaint should result in compensation.
17. The method of claim 15 , further comprising training the machine-learning predictive model using a plurality of historical complaints, wherein the trained machine-learning predictive model is executed to generating the prediction.
18. The method of claim 15 , wherein the machine-learning predictive model is a sequential neural network.
19. A method comprising:
obtaining a first set of complaint logs corresponding to one or more user complaints, each complaint log of the first set of complaint logs comprising a resolution comments field and an indication whether compensation was provided, the resolution comments field providing a textual representation of a resolution for a corresponding complaint log;
parsing the resolution comments field of each complaint log of the first set of complaint logs to identify a first set of key terms;
training a machine-learning predictive model using the first set of key terms, the machine-learning predictive model being a supervised machine learning model configured to accept the parsed resolution comments field for the first set of complaint logs as inputs and to output a probability, for each of the first set of complaint logs, that a corresponding user complaint should have compensation, the machine-learning model comprising a sequential neural network;
obtaining a second set of complaint logs corresponding to one or more additional user complaints;
parsing a resolution comments field of each complaint log of the second set of complaint logs to determine a frequency the each term of first set of key terms appears in the resolution comments fields of the second set of complaint logs; and
executing the trained machine-learning predictive model based on the second set of complaint logs to generate, for each of the second set of complaint logs, a probability that a corresponding user complaint should have compensation.