Interaction response guidance

1 . A system comprising processing circuitry and a storage device, wherein the processing circuitry has access to the storage device and is configured to:

identify, based on text derived from a message originating from a user device, a topic associated with the message, wherein the topic is identified from a plurality of topics;

generate, based on the topic and the text derived from the message, one or more productive expressions that, if communicated to the user device, tend to encourage a desired outcome;

generate, based on the topic and the text derived from the message, one or more unproductive expressions that, if communicated to the user device, tend to discourage the desired outcome; and

output a user interface that presents the topic associated with the message, the productive expressions, and the unproductive expressions, wherein to output the user interface, the processing circuitry is further configured to:

output the user interface to an agent device operated by an agent,

receive input indicating a response to the message, and

communicate, to the user device, the response.

2 . The system of claim 1 , wherein the processing circuitry is further configured to:

receive data representing the message from the user device; and

preprocess the data representing the message to generate the text derived from the message.

3 . The system of claim 2 , wherein the data representing the message is an audio signal, and wherein to preprocess the data, the processing circuitry is further configured to:

convert the audio signal into the text by applying a natural language processing model to the audio signal.

4 . The system of claim 1 , wherein to identify the topic associated with the message, the processing circuitry is further configured to:

apply a latent Dirichlet allocation model to the text associated with the message.

5 . The system of claim 1 , wherein to generate the one or more productive expressions, the processing circuitry is further configured to:

choose, based on the identified topic, a machine learning model from a plurality of topic-based machine learning models, wherein each of the topic-based machine learning models are associated with a different one of the plurality of topics; and

apply the chosen machine learning model to the text derived from the message.

6 . The system of claim 1 , wherein to generate the one or more productive expressions and to generate the one or more unproductive expressions, the processing circuitry is further configured to:

choose, based on the identified topic, a machine learning model from a plurality of topic-based machine learning models, wherein each of the topic-based machine learning models are associated with a different one of the plurality of topics; and

apply the chosen machine learning model to the text derived from the message.

7 . The system of claim 1 , wherein to receive input indicating the response, the processing circuitry is further configured to:

receive a proposed response from the agent device; and

apply a deep learning model to predict, based on the message and the proposed response, a predicted likelihood of achieving the desired outcome.

8 . The system of claim 7 , wherein to receive input indicating the response, the processing circuitry is further configured to:

output a feedback user interface to the agent device that presents the predicted likelihood of achieving the desired outcome, wherein the feedback user interface further indicates which words in the proposed response tend to encourage the desired outcome and which words in the proposed response tend to discourage the desired outcome; and

receive the response from the agent device, wherein the response is a modification of the proposed response generated based on the feedback user interface.

9 . The system of claim 1 , wherein the processing circuitry is further configured to:

retrain, based on the message and the response, at least one machine learning model from a plurality of topic-based machine learning models.

10 . The system of claim 1 ,

wherein the system is operated by a bank,

wherein the user device is operated by a user that is a customer of the bank, and

wherein the desired outcome is the bank retaining the user as a customer of the bank.

11 . The system of claim 1 , wherein the processing circuitry is further configured to:

receive a plurality of historical messages and responses, each labeled to indicate whether the desired outcome was achieved; and

apply a classification model to generate, based on the plurality of historical messages and responses, a set of expressions associated with the desired outcome being achieved and a set of topics and expressions associated with the desired outcome not being achieved.

12 . A method comprising:

identifying, by a system and based on text derived from a message originating from a user device, a topic associated with the message, wherein the topic is identified from a plurality of topics;

generating, by the system and based on the topic and the text derived from the message, one or more productive expressions that, if communicated to the user device, tend to encourage a desired outcome;

generating, by the system and based on the topic and the text derived from the message, one or more unproductive expressions that, if communicated to the user device, tend to discourage the desired outcome; and

outputting, by the system, a user interface that presents the topic associated with the message, the productive expressions, and the unproductive expressions, wherein outputting the user interface includes:

outputting, by the system, the user interface to an agent device operated by an agent,

receiving, by the system, input indicating a response to the message, and

communicating, by the system and to the user device, the response.

13 . The method of claim 12 , wherein the method further comprises:

receiving, by the system, data representing the message from the user device; and

preprocessing, by the system, the data representing the message to generate the text derived from the message.

14 . The method of claim 13 , wherein the data representing the message is an audio signal, and wherein preprocessing the data includes:

converting the audio signal into the text by applying a natural language processing model to the audio signal.

15 . The method of claim 12 , wherein identifying the topic associated with the message includes:

applying a latent Dirichlet allocation model to the text associated with the message.

16 . The method of claim 12 , wherein generating the one or more productive expressions includes:

choosing, based on the identified topic, a machine learning model from a plurality of topic-based machine learning models, wherein each of the topic-based machine learning models are associated with a different one of the plurality of topics; and

applying the chosen machine learning model to the text derived from the message.

17 . The method of claim 12 , wherein generating the one or more productive expressions and generating the one or more unproductive expressions includes:

choosing, based on the identified topic, a machine learning model from a plurality of topic-based machine learning models, wherein each of the topic-based machine learning models are associated with a different one of the plurality of topics; and

applying the chosen machine learning model to the text derived from the message.

18 . Non-transitory computer-readable media comprising instructions that, when executed, cause processing circuitry of a system to:

identify, based on text derived from a message originating from a user device, a topic associated with the message, wherein the topic is identified from a plurality of topics;

generate, based on the topic and the text derived from the message, one or more productive expressions that, if communicated to the user device, tend to encourage a desired outcome;

generate, based on the topic and the text derived from the message, one or more unproductive expressions that, if communicated to the user device, tend to discourage the desired outcome; and

output a user interface that presents the topic associated with the message, the productive expressions, and the unproductive expressions, wherein to output the user interface, the instructions further cause the processing circuitry to:

output the user interface to an agent device operated by an agent,

receive input indicating a response to the message, and

communicate, to the user device, the response.