Intent prediction for dialogue generation
In certain embodiments, intent prediction and dialogue generation may be facilitated. In some embodiments, a chat initiation request may be obtained from a user. The latest activity information associated with the user may be provided to a prediction model to obtain a first set of predicted intents of the user. For each intent of the first set of predicted intents, a candidate question may be selected from a question set based on the candidate question matching the intent. In some embodiments, the candidate questions may be simultaneously presented on the chat interface.
1. A system for facilitating chat dialogue, the system comprising: one or more processors programmed with computer program instructions that, when executed, cause operations comprising:
obtaining, via a neural network, predicted intents of a user based on activity information associated with the user;
prior to or during initiation of a chat session with the user, assigning the predicted intents to two or more sets of predicted intents based on probability categories associated with the predicted intents such that:
a first set of predicted intents comprises at least one predicted intent associated with each of the probability categories, each predicted intent being associated with a probability in a respective probability category; and
a second set of predicted intents comprises at least one predicted intent associated with each of the probability categories, each predicted intent being associated with a probability in a respective probability category; and
upon the initiation of the chat session with the user, causing questions corresponding to the first set of predicted intents to be presented on an interface based on a selection of the first set of predicted intents over the second set of predicted intents, wherein each question of the questions is selected from a set of questions for a corresponding predicted intent of the first set of predicted intents.
2. The system of claim 1 , wherein the activity information associated with the user is provided as input to the neural network responsive to a chat initiation request, and
wherein assigning the predicted intents comprises randomly selecting, from the predicted intents associated with a probability in a first category of the probability categories, at least one predicted intent to be assigned to the first set of predicted intents.
3. The system of claim 1 , the operations further comprising:
obtaining, via the interface, a user selection of a given question of the questions, the given question matching a given intent of the first set of predicted intents;
determining a feedback score associated with the given intent based on a probability category of the given intent such that (i) the feedback score is a first feedback score responsive to the given intent being in a first category and (ii) the feedback score is a second feedback score responsive to the given intent being in a second category; and
using, based on the user selection of the given question, the given intent and the feedback score to update one or more configurations of the neural network.
4. The system of claim 1 , wherein the activity information associated with the user comprises page view information related to recent page views of the user, service interaction information related to recent interactions of the user with one or more services, or transaction information related to recent transactions of the user.
5. A method comprising:
obtaining, via a prediction model, predicted intents of a user based on activity information associated with the user;
prior to or during initiation of a chat session with the user, assigning the predicted intents to two or more sets of predicted intents based on probability categories associated with the predicted intents such that:
a first set of predicted intents comprises at least one predicted intent associated with each of the probability categories, each of the predicted intents being associated with a probability in a respective probability category; and
a second set of predicted intents comprises at least one predicted intent associated with each of the probability categories, each of the predicted intents being associated with a probability in a respective probability category; and
upon the initiation of the chat session with the user, causing questions corresponding to the first set of predicted intents to be presented on an interface based on a selection of the first set of predicted intents over the second set of predicted intents, wherein each question of the questions is selected from a set of questions for a corresponding predicted intent of the first set of predicted intents.
6. The method of claim 5 , further comprising:
obtaining a chat initiation request from the user,
wherein the activity information associated with the user is provided as input to the prediction model responsive to the chat initiation request.
7. The method of claim 5 , wherein the first set of predicted intents comprises at least one predicted intent associated with a confidence score in a first score tier and at least one predicted intent associated with a confidence score in a second score tier lower than the first score tier, and wherein the second set of predicted intents comprises at least another predicted intent associated with a confidence score in the first score tier and at least another predicted intent associated with a confidence score in the second score tier.
8. The method of claim 5 , further comprising:
obtaining, via the interface, a user selection of a given question of the questions, the given question matching a given intent of the first set of predicted intents;
determining a feedback score associated with the given intent based on a probability category of the given intent such that (i) the feedback score is a first feedback score responsive to the given intent being in a first category and (ii) the feedback score is a second feedback score responsive to the given intent being in a second category; and
using, based on the user selection of the given question, the given intent and the feedback score to update one or more configurations of the prediction model.
9. The method of claim 5 , further comprising:
selecting the questions for the presentation on the interface by, for each intent of the predicted intents, selecting a question from a question set based on the question matching the intent.
10. The method of claim 5 , wherein assigning the predicted intents comprises:
randomly selecting, from the predicted intents associated with a first probability tier, at least one predicted intent to be assigned to the first set of predicted intents; or
randomly selecting, from the predicted intents associated with a second probability tier, at least one predicted intent to be assigned to the first set of predicted intents.
11. The method of claim 5 , wherein the activity information associated with the user comprises page view information related to recent page views of the user.
12. The method of claim 5 , wherein the activity information associated with the user comprises service interaction information related to recent interactions of the user with one or more services or transaction information related to recent transactions of the user.
13. The method of claim 5 , wherein the prediction model comprises a neural network.
14. A non-transitory computer-readable media comprising: instructions that, when executed, cause operations comprising:
obtaining, via a prediction model, predicted intents of a user and probability categories for a plurality of different categories based on activity information associated with the user;
prior to or during initiation of a chat session with the user, assigning the predicted intents to two or more sets of predicted intents based on the probability categories associated with the predicted intents such that:
a first set of predicted intents comprises at least one predicted intent associated with each of the categories, each predicted intent is associated with a probability in a respective probability category; and
a second set of predicted intents comprises at least one predicted intent associated with each of the categories, each predicted intent is associated with a probability in a respective probability category; and
upon the initiation of the chat session with the user, causing dialogue items corresponding to the first set of predicted intents to be presented on an interface based on a selection of the first set of predicted intents over the second set of predicted intents, wherein each dialogue item is selected from a set of questions for a corresponding predicted intent of the first set of predicted intents.
15. The media of claim 14 , wherein the first set of predicted intents comprises at least one predicted intent associated with a confidence score in a first score tier and at least one predicted intent associated with a confidence score in a second score tier lower than the first score tier, and wherein the second set of predicted intents comprises at least another predicted intent associated with a confidence score in the first score tier and at least another predicted intent associated with a confidence score in the second score tier.
16. The media of claim 14 , the operations further comprising:
obtaining, via the interface, a user selection of a given dialogue item of the dialogue items, the given dialogue item matching a given intent of the first set of predicted intents;
determining a feedback score associated with the given intent based on a probability category of the given intent such that (i) the feedback score is a first feedback score based on the given intent being in a first category and (ii) the feedback score is a second feedback score based on the given intent being in a second category; and
using, based on the user selection of the given dialogue item, the given intent and the feedback score to update one or more configurations of the prediction model.
17. The media of claim 14 , the operations further comprising:
selecting the dialogue items for the presentation on the interface by, for each intent of the predicted intents, selecting a dialogue item from a dialogue item set based on the dialogue item matching the intent.
18. The media of claim 14 , wherein assigning the predicted intents comprises:
randomly selecting, from the predicted intents associated with a probability in the first category of the probability categories, at least one predicted intent to be assigned to the first set of predicted intents; or
randomly selecting, from the predicted intents associated with a probability in the second category of the probability categories, at least one predicted intent to be assigned to the first set of predicted intents.
19. The media of claim 14 , wherein the activity information associated with the user comprises page view information related to page views of the user.
20. The media of claim 14 , wherein the prediction model comprises a neural network.