Empathetic query response using mixture of experts

1. A computer-implemented method comprising:

receiving, by a system operatively coupled to a processor, an input query comprising an emotion portion and a technical portion;

generating, by the system, a filtered input query by removing the emotion portion from the input query;

generating, by the system, using the filtered input query and a first machine learning model trained using training data having emotion portions removed, a first response portion that comprises only a technical response to the technical portion of the filtered input query;

generating, by the system, using the input query and a second machine learning model that employs in-context learning based on an emotion classification of the emotion portion, a second response portion that comprises only an emotional response to the emotion portion of the input query; and

combining, by the system, the first response portion and the second response portion into a final response.

2. The computer-implemented method of claim 1 , wherein the first machine learning model comprises a first generative pre-trained transformer (GPT) model that utilizes fine-tuning, and the second machine learning model comprises a second GPT model that utilizes the in-context learning.

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

training, by the system, the first machine learning model to output the first response portion based on the technical portion of the input query, wherein the training comprises:

collecting data of completed customer service requests;

transforming the data into the training data comprising question-and-answer pairs with the emotion portions removed; and

tuning the first machine learning model with the question-and-answer pairs.

4. The computer-implemented method of claim 1 , wherein the generating the second response portion comprises:

classifying the input query into at least one emotion class of a group of emotion classes;

selecting one or more emotion based prompts from a emotion based prompt bank based on emotion the at least one emotion class;

concatenating the one or more emotion based prompts and the input query into one or more respective concatenated prompts; and

inputting the one or more respective concatenated prompts and the input query to the second machine learning model.

5. The computer-implemented method of claim 4 , wherein the one or more emotion based prompts comprise pairs of a question having an emotional tone and an emotion based answer based on an emotion class of the emotional tone of the question.

6. The computer-implemented method of claim 4 , wherein the classifying comprises:

generating classification labels and confidence scores of the classification labels; and

categorizing the input query based on the classification labels and confidence scores.

7. The computer-implemented method of claim 4 , wherein the respective concatenated prompts do not have an answer for the technical portion of the input query.

8. A computer program product, the computer program product comprising a non-transitory computer readable medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

receive, by the processor, an input query comprising an emotion portion and a technical portion;

generate, by the processor, a filtered input query by removing the emotion portion from the input query;

generate, by the processor, using the filtered input query and a first machine learning model trained using training data having emotion portions removed, a first response portion that comprises only a technical response to the technical portion of the filtered input query;

generate, by the processor, using the input query and a second machine learning model that employs in-context learning based on an emotion classification of the emotion portion, a second response portion that comprises only an emotional response to the emotion portion of the input query; and

combine, by the processor, the first response portion and the second response portion into a final response.

9. The computer program product of claim 8 , wherein the first machine learning model comprises a first GPT model that utilizes fine-tuning, and the second machine learning model comprises a second GPT model that utilizes the in-context learning.

10. The computer program product of claim 8 , wherein the program instructions further cause the processor to:

train, by the processor, the first machine learning model to output the first response portion based on the technical portion of the input query, wherein the training comprises:

collecting data of completed customer service requests;

transforming the data into the training data comprising question-and-answer pairs with the emotion portions removed; and

tuning the first machine learning model with the question-and-answer pairs.

11. The computer program product of claim 8 , wherein the generating the second response portion comprises:

classifying the input query into at least one emotion class of a group of emotion classes;

selecting one or more emotion based prompts from a emotion based prompt bank based on emotion the at least one emotion class;

concatenating the one or more emotion based prompts and the input query into one or more respective concatenated prompts; and

inputting the concatenated prompts and the input query to the second machine learning model.

12. The computer program product of claim 11 , wherein the one or more emotion based prompts comprise pairs of a question having an emotional tone and an emotion based answer based on an emotion class of the emotional tone of the question.

13. The computer program product of claim 11 , wherein the classifying comprises:

generating classification labels and confidence scores of the classification labels; and

categorizing the input query based on the classification labels and confidence scores.

14. The computer program product of claim 11 , wherein the respective concatenated prompts do not have an answer for the technical portion of the input query.

15. A system comprising:

a memory that stores computer executable components; and

a processor, operably coupled to the memory, and that executes the computer executable components stored in the memory, wherein the computer executable components comprise:

a communications component that receives an input query comprising an emotion portion and a technical portion;

a filtering component that generates a filtered input query by removing the emotion portion from the input query;

a first machine learning model, trained using training data having emotion portions removed, that generates, using the filtered input query, a first response portion that comprises only a technical response to the technical portion of the filtered input query;

a second machine learning model, employing in-context learning based on an emotion classification of the emotion portion, that generates, using the input query, a second response portion that comprises only an emotional response to the emotion portion of the input query; and

a combination component that combines the first response portion and the second response portion into a final response.

16. The system of claim 15 , wherein the first machine learning model comprises a first GPT model that utilizes fine-tuning, and the second machine learning model comprises a second GPT model that utilizes the in-context learning.

17. The system of claim 15 , wherein the computer executable components further comprise a technical training component that trains the first machine learning model to output the first response portion based on the technical portion of the input query, wherein the training comprises:

collecting data of completed customer service requests;

transforming the data into the training data comprising question-and-answer pairs with the emotion portions removed; and

tuning the first machine learning model with the question-and-answer pairs.

18. The system of claim 15 , wherein the computer executable components further comprise an in-context learning component that:

classifies the input query into at least one emotion class of a group of emotion classes;

selects one or more emotion based prompts from a emotion based prompt bank based on emotion the at least one emotion class;

concatenates the one or more emotion based prompts and the input query into one or more respective concatenated prompts; and

inputs the one or more respective concatenated prompts and the input query to the second machine learning model.

19. The system of claim 18 , wherein the one or more emotion based prompts comprise pairs of a question having an emotional tone and an emotion based answer based on an emotion class of the emotional tone of the question.

20. The system of claim 18 , wherein the respective concatenated prompts do not have an answer for the technical portion of the input query.