Automated conversation review to surface virtual assistant misunderstandings
A scalable system provides automated conversation review that can identify potential miscommunications. The system may provide suggested actions to fix errors in intelligent virtual assistant (IVA) understanding, may prioritize areas of language model repair, and may automate the review of conversations. By the use of an automated system for conversation review, problematic interactions can be surfaced without exposing the entire set of conversation logs to human reviewers, thereby minimizing privacy invasion. A scalable system processes conversations and autonomously marks the interactions where the IVA is misunderstanding the user.
1. A method for automated conversation review to identify virtual assistant misunderstandings, comprising:
receiving conversation data between a user and an intelligent virtual assistant (IVA), wherein the conversation data comprises a plurality of user inputs;
identifying an intent from the conversation data, the intent being derived using a language model that maps the user inputs to the intent;
generating a risk score for each of the plurality of user inputs, wherein the risk score represents a risk of intent misclassification of each user input by detecting features of intent error and aggregating the features into the risk score;
determining, using the intent the risk score for each of the plurality of user inputs, and a binary classifier to perform automated voting, that the conversation data has a misunderstood intent; and
adjusting the language model responsive to determining that the conversation data has the misunderstood intent.
2. The method of claim 1 , wherein the determining that the conversation data has the misunderstood intent is performed without human review.
3. The method of claim 1 , wherein determining that the conversation data has the misunderstood intent comprises ranking the user inputs by risk score for priority review.
4. The method of claim 3 , further comprising reviewing the user inputs based on ranking to determine whether each reviewed turn was misunderstood.
5. The method of claim 1 , wherein adjusting the language model comprises adjusting the IVA.
6. The method of claim 1 , further comprising processing the conversation data with one or more natural language processing techniques to identify the intent for the conversation data.
7. The method of claim 6 , further comprising determining, based at least in part on a presence of one or more indicators of intent error for the conversation data, a measure of confidence that the intent is correctly identified for the user input, and associating the measure of confidence with the conversation data.
8. The method of claim 1 , further comprising:
receiving a selection of the intent to obtain feedback regarding the intent;
presenting a feedback interface that enables a voter to provide feedback regarding matching of the conversation data to the intent;
receiving feedback for the voter regarding an accuracy of matching the conversation data to the intent;
evaluating the intent based at least in part on the feedback;
determining that the feedback indicates that the matching of the conversation data to the intent is not accurate;
increasing a weighting to be applied to the one or more indicators of intent error based at least in part on the determining that the feedback indicates that the matching of the conversation data to the intent is not accurate; and
applying the weighting to the one or more indicators of intent error.
9. A computer system, comprising:
a processor; and
a memory storing instructions which when executed by the computer system causes the computer system to:
receive conversation data between a user and an intelligent virtual assistant (IVA), wherein the conversation data comprises a plurality of user inputs;
identify an intent from the conversation data, the intent being derived using a language model that maps the user inputs to the intent;
generate a risk score for each of the plurality of user inputs, wherein the risk score represents a risk of intent misclassification of each user input by detecting features of intent error and aggregating the features into the risk score;
determine, using the intent the risk score for each of the plurality of user inputs, and a binary classifier to perform automated voting, that the conversation data has a misunderstood intent; and
adjust the language model responsive to determining that the conversation data has the misunderstood intent.
10. The computer system of claim 9 , further comprising instructions that cause the computer system to rank the user inputs by risk score for priority review.
11. The computer system of claim 9 , further comprising instructions that cause the computer system to adjust the IVA.
12. The computer system of claim 9 , further comprising instructions that cause the computer system to process the conversation data with one or more natural language processing techniques to identify the intent for the conversation data.
13. The computer system of claim 12 , further comprising instructions that cause the computer system to determine, based at least in part on a presence of one or more indicators of intent error for the conversation data, a measure of confidence that the intent is correctly identified for the user input, and associating the measure of confidence with the conversation data.
14. The computer system of claim 9 , further comprising instructions that cause the computer system to:
receive a selection of the intent to obtain feedback regarding the intent;
present a feedback interface that enables a voter to provide feedback regarding matching of the conversation data to the intent;
receive feedback for the voter regarding an accuracy of matching the conversation data to the intent;
evaluate the intent based at least in part on the feedback;
determine that the feedback indicates that the matching of the conversation data to the intent is not accurate;
increase a weighting to be applied to the one or more indicators of intent error based at least in part on the determining that the feedback indicates that the matching of the conversation data to the intent is not accurate; and
apply the weighting to the one or more indicators of intent error.
15. A non-transitory computer-readable storage medium comprising instructions that, when executed by one or more processors of a computing system, cause the computing system to perform a method for automated conversation review to identify virtual assistant misunderstandings, the non-transitory computer-readable storage medium comprising instructions to:
receive conversation data between a user and an intelligent virtual assistant (IVA), wherein the conversation data comprises a plurality of user inputs;
identify an intent from the conversation data, the intent being derived using a language model that maps the user inputs to the intent;
generate a risk score for each of the plurality of user inputs, wherein the risk score represents a risk of intent misclassification of each user input by detecting features of intent error and aggregating the features into the risk score;
determine, using the intent the risk score for each of the plurality of user inputs and a binary classifier to perform automated voting, that the conversation data has a misunderstood intent; and
adjust the language model responsive to determining that the conversation data has the misunderstood intent.
16. The non-transitory computer-readable storage medium of claim 15 , further comprising instructions to rank the user inputs by risk score for priority review.
17. The non-transitory computer-readable storage medium of claim 15 , further comprising instructions to determine, based at least in part on a presence of one or more indicators of intent error for the conversation data, a measure of confidence that the intent is correctly identified for the user input, and associating the measure of confidence with the conversation data.
18. The non-transitory computer-readable storage medium of claim 15 , further comprising instructions to:
receive a selection of the intent to obtain feedback regarding the intent;
present a feedback interface that enables a voter to provide feedback regarding matching of the conversation data to the intent;
receive feedback for the voter regarding an accuracy of matching the conversation data to the intent;
evaluate the intent based at least in part on the feedback;
determine that the feedback indicates that the matching of the conversation data to the intent is not accurate;
increase a weighting to be applied to the one or more indicators of intent error based at least in part on the determining that the feedback indicates that the matching of the conversation data to the intent is not accurate; and
apply the weighting to the one or more indicators of intent error.