Granted Patent B1

US 12,585,683 · App. 18/893,660 · Granted Mar 24, 2026

Inventors: Jun Su (Beijing, CN); Yin Xia (Beijing, CN); Douglas Brian Mueller (Apex, NC); Peng Hui Jiang (Beijing, CN)

Assignee: International Business Machines Corporation

G06F16/3337G06F16/3347G06F16/383

Abstract

Claims (80)

1 . A method comprising:

causing a received query to be translated from a first language to a second language;

causing potential answers to be generated for the translated query;

causing features to be extracted from the translated query;

causing the extracted features to be converted into feature vectors;

causing the feature vectors to be compared against existing vectors in a knowledge base that correspond to past question-answer pairs;

causing the potential answers to be ranked based at least in part on an outcome of comparing the feature vectors against the existing vectors in the knowledge base;

causing a subset of the potential answers to be enriched with additional context and/or information;

causing a final answer to be generated based at least in part on the ranked potential answers and the enriched potential answers by:

combining at least one of the potential answers which addresses at least a portion of the received query and does not match the past question-answer pairs in the knowledge base, and two or more of the potential answers and/or the enriched potential answers which address portions of the received query and match respective entries in the knowledge base;

using an AI based model to apply weight values to feedback received from a user in response to the final answer being shown on a display screen; and

using the weighted feedback to re-train the AI based model.

2 . The method of claim 1 , wherein the causing the final answer to be generated includes:

evaluating correctness and context of the ranked potential answers and the enriched potential answers.

3 . The method of claim 1 , wherein causing a subset of the potential answers to be enriched with additional context and/or information includes:

incorporating how often the respective potential answers have been modified in response to being generated as previous final answers.

4 . The method of claim 1 , further comprising:

causing the final answer to be translated back to the first language; and

solving the received query by outputting the translated final answer,

wherein at least some of the past question-answer pairs in the knowledge base are expressed in languages that are different than the first and second languages.

5 . The method of claim 1 , wherein the knowledge base is formed by:

analyzing the past question-answer pairs;

extracting features from the past question-answer pairs; and

using the extracted features to build a historical question-answer model configured to capture characteristics of diverse language interactions.

6 . The method of claim 5 , further comprising:

converting the features extracted from the past question-answer pairs into the existing vectors; and

using the existing vectors to form a vector database in the knowledge base.

7 . The method of claim 1 , further comprising:

causing the final answer to be shown on the display screen.

8 . A computer program product comprising:

one or more computer-readable storage media; and

program instructions stored on the one or more storage media to perform operations comprising:

causing a received query to be translated from a first language to a second language;

causing potential answers to be generated for the translated query;

causing features to be extracted from the translated query;

causing the extracted features to be converted into feature vectors;

causing the feature vectors to be compared against existing vectors in a knowledge base that correspond to past question-answer pairs;

causing the potential answers to be ranked based at least in part on an outcome of comparing the feature vectors against the existing vectors in the knowledge base;

causing a subset of the potential answers to be enriched with additional context and/or information by modifying the potential answers in the subset to incorporate how often the respective potential answers have been modified in response to being generated as previous final answers;

evaluating correctness and context of the ranked potential answers and the enriched potential answers;

causing a final answer to be generated based at least in part on the ranked potential answers and the enriched potential answers; and

using feedback that has been weighted to re-train an AI based model, wherein the unweighted feedback is received from a user in response to the final answer being shown on a display screen.

9 . The computer program product of claim 8 , wherein the causing the final answer to be generated based at least in part on the ranked potential answers and the enriched potential answers includes:

combining at least one of the potential answers which addresses at least a portion of the received query and does not match the past question-answer pairs in the knowledge base, and two or more of the potential answers and/or the enriched potential answers which address portions of the received query and match respective entries in the knowledge base.

10 . The computer program product of claim 8 , wherein at least some of the past question-answer pairs in the knowledge base are expressed in languages that are different than the first and second languages.

11 . The computer program product of claim 8 , wherein the operations further comprise:

causing the final answer to be translated back to the first language; and

solving the received query by outputting the translated final answer.

12 . The computer program product of claim 8 , wherein the causing the potential answers to be ranked based at least in part on the outcome of comparing the feature vectors against the existing vectors in the knowledge base, includes:

using a similarity scoring mechanism to determine a cosine similarity between the received query and the question-answer pairs in the knowledge base.

13 . The computer program product of claim 8 , wherein the knowledge base is formed by:

analyzing the past question-answer pairs;

extracting features from the past question-answer pairs; and

using the extracted features to build a historical question-answer model configured to capture characteristics of diverse language interactions.

14 . The computer program product of claim 13 , wherein the operations further comprise:

converting the features extracted from the past question-answer pairs into the existing vectors; and

using the existing vectors to form a vector database in the knowledge base.

15 . A computer system comprising:

a processor set;

one or more computer-readable storage media; and

program instructions stored on the one or more storage media to cause the processor set to perform operations comprising:

causing a received query to be translated from a first language to a second language;

causing potential answers to be generated for the translated query;

causing features to be extracted from the translated query;

causing the extracted features to be converted into feature vectors;

causing the feature vectors to be compared against existing vectors in a knowledge base that correspond to past question-answer pairs;

causing the potential answers to be ranked based at least in part on an outcome of comparing the feature vectors against the existing vectors in the knowledge base;

causing a subset of the potential answers to be enriched by altering the potential answers in the subset based at least in part on how the respective potential answers have been modified in response to being generated as previous final answers;

evaluating correctness and context of the ranked potential answers and the enriched potential answers;

causing a final answer to be generated based at least in part on the ranked potential answers and the enriched potential answers; and

using feedback that has been weighted to re-train an AI based model, wherein the unweighted feedback is received from a user in response to the final answer being shown on a display screen.

16 . The computer system of claim 15 , wherein the causing a final answer to be generated based at least in part on the ranked potential answers and the enriched potential answers includes:

combining at least one of the potential answers which addresses at least a portion of the received query and does not match the past question-answer pairs in the knowledge base, and two or more of the potential answers and/or the enriched potential answers which address portions of the received query and match respective entries in the knowledge base.

17 . The computer system of claim 15 , wherein at least some of the past question-answer pairs in the knowledge base are expressed in languages that are different than the first and second languages.

18 . The computer system of claim 15 , wherein the knowledge base is formed by:

analyzing the past question-answer pairs;

extracting features from the past question-answer pairs;

using the extracted features to build a historical question-answer model configured to capture characteristics of diverse language interactions;

converting the features extracted from the past question-answer pairs into the existing vectors; and

using the existing vectors to form a vector database in the knowledge base.

References Cited (23)