Granted Patent B2

US 12,731,582 · App. 18/112,858 · Granted Sep 8, 2026

Inventors: Michelle Brachman (Quincy, MA); James Johnson (Somerville, MA); Qian Pan (Canton, MA); Casey Dugan (Cambridge, MA)

Assignee: International Business Machines Corporation

G10L15/22G10L15/07G10L2015/223

Abstract

Claims (68)

1 . A computer-implemented method of analyzing a user input via natural language processing comprising:

training a machine learning based natural language processing (NLP) model, using sequences of tasks, to determine one or more computing tasks from a constrained set of computing tasks based at least on a knowledge base and user utterance data;

analyzing, using the machine learning based NLP model, a natural language utterance from a user to determine the one or more computing tasks;

analyzing the natural language utterance to identify one or more common n-grams, from prior attempt records of the knowledge base, that are similar to the natural language utterance;

identifying, based on the one or more common n-grams, one or more modifications to the natural language utterance that are based on previous user modifications to a previous user utterance;

receiving an indication that the user accepted at least one modification of the one or more modifications, wherein the at least one modification modifies the one or more computing tasks and the one or more modifications are selected based on a score of each of a plurality of previous user utterances, and wherein the score is determined, using the prior attempt records, based on how many times a previous user accepted the at least one modification;

executing, using the machine learning based NLP model, the modified one or more computing tasks;

updating the knowledge base based at least in part on executing the modified one or more computing tasks; and

retraining the machine learning based NLP model using the updated knowledge base.

2 . The computer-implemented method of claim 1 ,

wherein the user accepting the at least one modification adjusts a score of the previous user utterance in the knowledge base.

3 . The computer-implemented method of claim 1 , further comprising:

prompting the user via a user interface that displays text comprising the natural language utterance of the user and one or more visual elements corresponding to the one or more modifications.

4 . The computer-implemented method of claim 1 ,

wherein the knowledge base is updated in response to the user providing a previously-unprompted modification.

5 . The computer-implemented method of claim 1 ,

wherein the knowledge base includes a plurality of records of previous user interactions of a plurality of users, wherein each previous user interactions includes at least one initial utterance and at least one modified utterance.

6 . The computer-implemented method of claim 1 ,

wherein the one or more modifications include a word or phrase to be added to the natural language utterance, to be replaced in the natural language utterance, or to be removed from the natural language utterance.

7 . A computer system for analyzing user input via natural language processing comprising:

one or more processors; and

one or more memory devices coupled to the one or more processors, wherein the one or more processors are configured to:

train a machine learning based natural language processing (NLP) model, using sequences of tasks, to determine one or more computing tasks from a constrained set of computing tasks based at least on a knowledge base and user utterance data;

analyze, using the machine learning based NLP model, a natural language utterance from a user to determine the one or more computing tasks;

analyze the natural language utterance to identify one or more common n-grams, from prior attempt records of the knowledge base, that are similar to the natural language utterance;

identify, based on the one or more common n-grams, one or more modifications to the natural language utterance that are based on previous user modifications to a previous user utterance;

receive an indication that the user accepted at least one modification of the one or more modifications, wherein the at least one modification modifies the one or more computing tasks and the one or more modifications are selected based on a score of each of a plurality of previous user utterances, and wherein the score is determined, using the prior attempt records, based on how many times a previous user accepted the at least one modification;

execute, using the machine learning based NLP model, the modified one or more computing tasks;

update the knowledge base based at least in part on executing the modified one or more computing tasks; and

retrain the machine learning based NLP model using the updated knowledge base.

8 . The computer system of claim 7 ,

wherein the user accepting the at least one modification adjusts a score of the previous user utterance in the knowledge base.

9 . The computer system of claim 7 ,

wherein the one or more processors are further configured to:

prompt the user via a user interface that displays text comprising the natural language utterance of the user and one or more visual elements corresponding to the one or more modifications.

10 . The computer system of claim 7 ,

wherein the knowledge base is updated in response to the user providing a previously-unprompted modification.

11 . The computer system of claim 7 ,

wherein the knowledge base includes a plurality of records of previous user interactions of a plurality of users, wherein each previous user interactions includes at least one initial utterance and at least one modified utterance.

12 . The computer system of claim 7 ,

wherein the one or more modifications include a word or phrase to be added to the natural language utterance, to be replaced in the natural language utterance, or to be removed from the natural language utterance.

13 . A non-transitory computer-readable medium storing a set of instructions for analyzing user input via natural language processing, the set of instructions comprising:

one or more instructions that, when executed by one or more processors of a device, cause the device to:

train a machine learning based natural language processing (NLP) model, using sequences of tasks, to determine one or more computing tasks from a constrained set of computing tasks based at least on a knowledge base and user utterance data;

analyze, using the machine learning based NLP model, a natural language utterance from a user to determine the one or more computing tasks;

analyze the natural language utterance to identify one or more common n-grams, from prior attempt records of the knowledge base, that are similar to the natural language utterance;

identify, based on the one or more common n-grams, one or more modifications to the natural language utterance that are based on previous user modifications to a previous user utterance;

receive an indication that the user accepted at least one modification of the one or more modifications, wherein the at least one modification modifies the one or more computing tasks and the one or more modifications are selected based on a score of each of a plurality of previous user utterances, and wherein the score is determined, using the prior attempt records, based on how many times a previous user accepted the at least one modification; and

execute, using the machine learning based NLP model, the modified one or more computing tasks;

update the knowledge base based at least in part on executing the modified one or more computing tasks; and

retrain the machine learning based NLP model using the updated knowledge base.

14 . The non-transitory computer-readable medium of claim 13 ,

wherein the user accepting the at least one modification adjusts a score of the previous user utterance in the knowledge base.

15 . The non-transitory computer-readable medium of claim 13 ,

wherein the one or more instructions cause the device to:

prompt the user via a user interface that displays text comprising the natural language utterance of the user and one or more visual elements corresponding to the one or more modifications.

16 . The non-transitory computer-readable medium of claim 13 ,

wherein the knowledge base is updated in response to the user providing a previously-unprompted modification.

17 . The non-transitory computer-readable medium of claim 13 ,

wherein the knowledge base includes a plurality of records of previous user interactions of a plurality of users, wherein each of the previous user interactions includes at least one initial utterance and at least one modified utterance.

18 . The computer-implemented method of claim 1 , further comprising:

vectorizing user utterances from the prior attempt records to identify an utterance closest to the natural language utterance of the user.

19 . The computer system of claim 7 ,

wherein the one or more processors are further configured to:

vectorize user utterances from the prior attempt records to identify an utterance closest to the natural language utterance of the user.

20 . The non-transitory computer-readable medium of claim 13 ,

wherein the one or more instructions cause the device to:

vectorize user utterances from the prior attempt records to identify an utterance closest to the natural language utterance of the user.

Assignments (1)

Continuity (1)

References Cited (19)