Evaluating conversation data based on risk factors

• US 10,545,648 B2
• Filed: 01/08/2019
• Issued: 01/28/2020
• Est. Priority Date: 09/09/2014
• Status: Active Grant

• Current Assignee: Verint Americas Incorporated (Verint Systems Incorporated)
• Original Assignee: Verint Americas Incorporated (Verint Systems Incorporated)
• Inventors: Beaver, Ian, Brown, Fred, Gossard, Casey
• Primary Examiner(s): Hang, Vu B
• Application Number:
  US16/242,639
• Publication Number:
  US 20190138190A1
• Time in Patent Office: 385 Days
• CPC Class Codes: G06F 3/04842 Selection of displayed objects or displayed text elements G06F3/0482 takes precedence
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 25/51 for comparison or discrimination
• CPC Class Codes: G10L 25/63 for estimating an emotional state
• CPC Class Codes: G10L 25/66 for extracting parameters related to health condition detecting or measuring for diagnostic purposes A61B5/00

Abstract

This disclosure describes techniques and architectures for evaluating conversations. In some instances, conversations with users, virtual assistants, and others may be analyzed to identify potential risks within a language model that is employed by the virtual assistants and other entities. The potential risks may be evaluated by administrators, users, systems, and others to identify potential issues with the language model that need to be addressed. This may allow the language model to be improved and enhance user experience with the virtual assistants and others that employ the language model.

Claims

1. One or more non-transitory computer-readable media storing computer-readable instructions that, when executed, instruct one or more processors to perform acts comprising:

causing a conversation user interface to be displayed on a display device to enable a conversation between a user and a virtual assistant;

receiving, via the conversation user interface, user input;

processing the user input with one or more natural language processing techniques to identify an intent unit for the user input, the intent unit being associated with a language model for the natural language processing techniques;

determining, based at least in part on a presence of one or more risk indicators for the user input, a measure of confidence that the intent unit is correctly identified for the user input;

associating the measure of confidence with the user input;

determining a health status of the intent unit based at least in part on the measure of confidence, the health status indicating a level of risk associated with the intent unit;

presenting the health status of the intent unit via an output device associated with an administrator;

receiving a selection of the intent unit to obtain feedback regarding the intent unit;

presenting a feedback interface that enables a voter to provide feedback regarding matching of the user input to the intent unit;

receiving feedback for the voter regarding an accuracy of matching the user input to the intent unit; and

evaluating the intent unit based at least in part on the feedback;

determining that the feedback indicates that the matching of the user input to the intent unit is not accurate;

increasing a weighting to be applied to the one or more risk indicators based at least in part on the determining that the feedback indicates that the matching of the user input to the intent unit is not accurate; and

applying the weighting to the one or more risk indicators;

wherein the determining the measure of confidence includes determining the measure of confidence based at least in part on the presence of the one or more weighted risk indicators.

2. The one or more non-transitory computer-readable media of claim 1, wherein the intent unit is associated with (i) an action to be performed at least partly by the virtual assistant and (ii) a pattern of components for triggering the intent unit.

3. The one or more non-transitory computer-readable media of claim 1, wherein the acts further comprise:

ranking the intent unit and another intent unit based at least in part on the health status of the intent unit and a health status of the other intent unit; and

presenting the ranking via the output device.

4. The one or more non-transitory computer-readable media of claim 1, wherein the one or more risk indicators include a plurality of risk indicators;

and the acts further comprise;

applying a weighting to each of the plurality of risk indicators to generate weighted risk indicators;

wherein the determining the measure of confidence includes determining the measure of confidence based at least in part on the weighted risk indicators.

5. The one or more computer-readable media of claim 1, wherein the feedback includes a vote indicating whether or not the user input matches the intent unit.

6. One or more non-transitory computer-readable media storing computer-readable instructions that, when executed, instruct one or more processors to perform acts comprising:

causing a conversation user interface to be displayed on a display device to enable a conversation between a user and a virtual assistant;

receiving, via the conversation user interface, user input;

processing the user input with one or more natural language processing techniques to identify an intent unit for the user input, the intent unit being associated with a language model for the natural language processing techniques;

determining, based at least in part on a presence of one or more risk indicators for the user input, a measure of confidence that the intent unit is correctly identified for the user input;

associating the measure of confidence with the user input;

determining a health status of the intent unit based at least in part on the measure of confidence, the health status indicating a level of risk associated with the intent unit;

presenting the health status of the intent unit via an output device associated with an administrator;

receiving a selection of the intent unit to obtain feedback regarding the intent unit;

presenting a feedback interface that enables a voter to provide feedback regarding matching of the user input to the intent unit;

receiving feedback for the voter regarding an accuracy of matching the user input to the intent unit; and

evaluating the intent unit based at least in part on the feedback;

determining that the feedback indicates that the matching of the user input to the intent unit is accurate;

decreasing a weighting to be applied to the one or more risk indicators based at least in part the on determining that the feedback indicates that the matching of the user input to the intent unit is accurate; and

applying the weighting to the one or more risk indicators;

wherein the determining the measure of confidence includes determining the measure of confidence based at least in part on the presence of the one or more weighted risk indicators.

7. A method comprising:

under control of a computing device configured with executable instructions,

identifying one or more risk factors to evaluate conversation data associated with one or more users, the conversation data representing at least one conversation;

processing the conversation data with a natural language processing system to identify an intent unit for the conversation data;

determining a confidence value for the conversation data based at least in part on the one or more risk factors, the confidence value indicating a level of confidence that the intent unit is accurately identified for the conversation data; and

wherein the one or more risk factors include a plurality of risk factors; and

the determining includes;

applying a weighting each of the plurality of risk factors; and

determining the confidence value based at least in part on the plurality of weighted risk factors;

determining a health status of the intent unit based at least in part on the measure of confidence, the health status indicating a level of risk associated with the intent unit;

presenting the health status of the intent unit via an output device associated with an administrator;

receiving a selection of the intent unit to obtain feedback regarding the intent unit;

presenting a feedback interface that enables a voter to provide feedback regarding matching of the user input to the intent unit;

receiving feedback for the voter regarding an accuracy of matching the user input to the intent unit; and

evaluating the intent unit based at least in part on the feedback;

determining that the feedback indicates that the matching of the user input to the intent unit is accurate;

decreasing a weighting to be applied to the one or more risk indicators based at least in part the on determining that the feedback indicates that the matching of the user input to the intent unit is accurate; and

applying the weighting to the one or more risk indicators;

wherein the determining the measure of confidence includes determining the measure of confidence based at least in part on the presence of the one or more weighted risk indicators.

8. The method of claim 7, further comprising utilizing the confidence value by:

determining that multiple pieces of user input are each associated with a confidence value that is below a threshold amount, the conversation data including the multiple pieces of user input;

determining a correlation between a user characteristic and the multiple pieces of user input; and

presenting the correlation via an output device.

9. The method of claim 8, wherein the user characteristic includes a demographic characteristic of a user.

10. The method of claim 8, wherein the user characteristic includes context that is used to interpret user input of the conversation data.

11. The method of claim 10, wherein the context includes at least one of a type of device that is used by a user to provide the user input, purchase history, or content that has been viewed by the user.

12. The method of claim 7, wherein the at least one conversation is between a user and a virtual assistant.

13. The method of claim 7, wherein the at least one conversation is between two or more users.

14. The method of claim 7, wherein the one or more risk factors include:

user feedback from a user regarding an evaluation of at least a portion of the least one conversation;

user feedback from an administrator regarding an evaluation of at least a portion of the at least one conversation;

a tone of a user'"'"'s voice during the at least one conversation;

a gesture of a user during the at least one conversation;

a facial expression of a user during the at least one conversation;

a sensor signal obtained from monitoring user response;

a message from a user about the at least one conversation;

or

a confidence value determined for the at least one conversation from at least one of a probabilistic model or statistical model.

15. The method of claim 7, wherein the at least one conversation includes user input;

and the one or more risk factors indicate;

whether or not the processing identifies an intent for the user input;

whether or not the user input directly preceded other user input for which the natural language processing system failed to identify an intent;

whether or not the user input is involved in a conversation in which the natural language processing system failed to identify an intent;

whether or not the user input is one of multiple user inputs for which the natural language processing system identifies a same intent two or more times in a row;

whether or not the user input is involved in a conversation in which the natural language processing system identifies a same intent two or more times in a row;

or

whether or not the user input is involved in a conversation in which the natural language processing system identifies a same intent two or more times.

16. The method of claim 7, wherein the one or more risk factors include a plurality of risk factors;

and the determining includes;

applying a weighting each of the plurality of risk factors; and

determining the confidence value based at least in part on the plurality of weighted risk factors.

17. The method of claim 7, further comprising:

receiving the conversation data during a conversation with a user;

wherein utilizing the confidence value includes utilizing the confidence value during the at least one conversation to determine whether or not an action that is associated with the intent should be performed.

18. The method of claim 7, wherein the utilizing the confidence value includes utilizing the confidence value to update a language model that is associated with the natural language processing system.

19. The method of claim 7, further comprising:

determining a health status based at least in part on the measure of confidence, the health status indicating a level of risk associated with the intent;

ranking the intent and another intent based at least in part on the health status of the intent and the health status of the other intent; and

presenting the ranking.

20. A system comprising:

one or more processors;

a database storing (i) conversation data representing at least one conversation and (ii) one or more risk factors;

memory communicatively coupled to the one or more processors and storing executable instructions that, when executed by the one or more processors, cause the one or more processors to perform acts comprising;

processing the conversation data with a natural language processing system to identify an intent for the conversation data;

determining a confidence value for the conversation data based at least in part on the one or more risk factors, the confidence value indicating a level of confidence that the intent is accurately identified for the conversation data; and

utilizing the confidence value;

wherein the acts further comprise;

presenting a feedback interface that enables a voter to vote regarding matching of the user input to the intent;

determining a profile factor for the voter based at least in part on a comparison of a previous number of votes from the voter to votes from a majority of voters, the majority of voters including more than a predetermined number of voters; and

applying a weighting to the vote based at least in part on the profile factor of the voter.