Granted Patent B1

US 11,341,335 · App. 16/741,630 · Granted May 24, 2022

Inventors: Paul Anthony Crook (Newcastle, WA); Baiyang Liu (Bellevue, WA); Pararth Paresh Shah (Sunnyvale, CA); Bing Liu (Sunnyvale, CA)

Assignee: Facebook Technologies, LLC

G06F40/35G06F3/167G06F9/453G06F9/547G06F16/3329G06K9/6221H04L51/32

Abstract

Claims (40)

1. A method comprising, by one or more computing systems:

receiving, from a client system associated with a user, a user input;

determining a task based on the user input, wherein the task is associated with one or more intents and one or more slots determined from the user input;

calculating a confidence score associated with the task based on one or more intent-confidence scores associated with the one or more intents, respectively, and one or more slot-confidence scores associated with the one or more slots, respectively;

generating one or more first dialog acts based on a task policy, wherein the task policy specifies dialog acts associated with the task;

generating, responsive to the confidence score being less than a threshold score, one or more second dialog acts by accessing an override policy, wherein the override policy specifies dialog acts that modify dialog acts specified by the task policy, and wherein the one or more second dialog acts are conditioned on the one or more first dialog acts; and

sending, to the client system, instructions for presenting a response to the user input, wherein the response is based on one or more of the first dialog acts or the second dialog acts.

2. The method of claim 1 , further comprising:

accessing a current dialog state associated with the user; and

updating the current dialog state based on the user input and contextual information associated with the user input.

3. The method of claim 2 , wherein generating the one or more first dialog acts is further based on the updated dialog state associated with the user.

4. The method of claim 1 , wherein one or more of the second dialog acts are associated with augmentations to one or more of the first dialog acts.

5. The method of claim 1 , wherein one or more of the second dialog acts are associated with replacements of one or more of the first dialog acts.

6. The method of claim 1 , wherein one or more of the second dialog acts are associated with one or more confirmation requests.

7. The method of claim 1 , wherein one or more of the second dialog acts are associated with terminations of one or more of the first dialog acts.

8. The method of claim 1 , wherein one or more of the second dialog acts are associated with a change of a modality associated with the instructions for presenting the response.

9. The method of claim 1 , wherein one or more of the second dialog acts are associated with a change of a user interface associated with the instructions for presenting the response.

10. The method of claim 1 , wherein one or more of the second dialog acts are associated with error-corrections of one or more of the first dialog acts.

11. The method of claim 1 , wherein generating the one or more first dialog acts comprises providing the user input to a third-party agent via an application programming interface (API), and wherein the one or more first dialog acts are generated based on an execution of the task based on the user input by the third-party agent.

12. The method of claim 1 , wherein generating the one or more second dialog acts further comprises:

analyzing the task policy and the one or more first dialog acts by one or more rules.

13. The method of claim 1 , wherein generating the one or more second dialog acts further comprises:

analyzing the task policy and the one or more first dialog acts by one or more machine-learning models, wherein the machine-learning models comprise one or more of a sequence-to-sequence model, a variational auto-encoder, a deep recurrent and convolutional neural network, or a generate-and-rank model.

14. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

receive, from a client system associated with a user, a user input;

determine a task based on the user input, wherein the task is associated with one or more intents and one or more slots determined from the user input;

calculate a confidence score associated with the task based on one or more intent-confidence scores associated with the one or more intents, respectively, and one or more slot-confidence scores associated with the one or more slots, respectively;

generate one or more first dialog acts based on a task policy, wherein the task policy specifies dialog acts associated with the task;

generate, responsive to the confidence score being less than a threshold score, one or more second dialog acts by accessing an override policy, wherein the override policy specifies dialog acts that modify dialog acts specified by the task policy, and wherein the one or more second dialog acts are conditioned on the one or more first dialog acts; and

send, to the client system, instructions for presenting a response to the user input, wherein the response is based on one or more of the first dialog acts or the second dialog acts.

15. The media of claim 14 , wherein one or more of the second dialog acts are associated with augmentations to one or more of the first dialog acts.

16. The media of claim 14 , wherein one or more of the second dialog acts are associated with replacements of one or more of the first dialog acts.

17. The media of claim 14 , wherein one or more of the second dialog acts are associated with one or more confirmation requests.

18. A system comprising: one or more processors; and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:

receive, from a client system associated with a user, a user input;

determine a task based on the user input, wherein the task is associated with one or more intents and one or more slots determined from the user input;

calculate a confidence score associated with the task based on one or more intent-confidence scores associated with the one or more intents, respectively, and one or more slot-confidence scores associated with the one or more slots, respectively;

generate one or more first dialog acts based on a task policy, wherein the task policy specifies dialog acts associated with the task;

generate, responsive to the confidence score being less than a threshold score, one or more second dialog acts by accessing an override policy, wherein the override policy specifies dialog acts that modify dialog acts specified by the task policy, and wherein the one or more second dialog acts are conditioned on the one or more first dialog acts; and

send, to the client system, instructions for presenting a response to the user input, wherein the response is based on one or more of the first dialog acts or the second dialog acts.

Assignments (3)

Cited By (9)

US 12,518,323 US 12,561,032 US 12,567,404 US 12,573,391 US 12,592,017 US 12,625,557 US 12,632,664 US 12,640,150 US 12,670,671