Graph and vector usage for automated QA system

Patent metadata

Field	Value
Publication number	US-12346356-B2
Application number	US-202318500761-A
Country	US
Kind code	B2
Filing date	Nov 2, 2023
Priority date	Nov 2, 2023
Publication date	Jul 1, 2025
Grant date	Jul 1, 2025

Abstract

A method for conversational query resolution includes receiving a query input from a user. The query input is decomposed into a plurality of tasks. A knowledge graph is queried to identify one or more relevant entities based on at least one of the plurality of tasks. A vector database is searched to identify one or more text chunks that correspond to the one or more relevant entities. Content relevant to at least one of the plurality of tasks is identified from the one or more text chunks. An answer is generated based on the identified content.

First claim

What is claimed is:

1. A computer-implemented method comprising:
receiving a query input from a user;
decomposing the query input into a plurality of tasks;
querying a knowledge graph to identify one or more relevant entities based on at least one of the plurality of tasks, wherein each of the one or more entities comprises text information related to a respective portion of a document;
searching a vector database to identify one or more text chunks, wherein each of the one or more text chunks comprises contextual information in natural language for a respective identified relevant entity and is linked to the respective identified relevant entity in the knowledge graph using a chunk identifier;
identifying content relevant to at least one of the plurality of tasks by parsing the contextual information in natural language within the one or more text chunks; and
generating an answer for the query based on the identified content.

2. The computer-implemented method of claim 1 , wherein each of the one or more text chunks comprises text extracted from a corpus of unstructured documents.

3. The computer-implemented method of claim 1 , wherein each respective text chunk is vectorized into a respective vector representation using an embedding model.

4. The computer-implemented method of claim 1 , wherein the content relevant to the plurality of tasks from the one or more text chunks is identified using a vector-based similarity technique.

5. The computer-implemented method of claim 4 , wherein the vector-based similarity technique used to identify the content comprises comparing at least one of (i) cosine similarity or (ii) distance similarity between vector representations of the one or more text chunks and each of the plurality of tasks.

6. The computer-implemented method of claim 1 , further comprising:
receiving feedback from the user regarding an accuracy of the answer; and
updating the knowledge graph based on the feedback.

7. The computer-implemented method of claim 6 , wherein updating the knowledge graph and the vector database based on the feedback comprises adding one or more new entities to the knowledge graph, wherein each of the one or more new entities comprises data derived from the answer.

8. The computer-implemented method of claim 1 , wherein generating the answer based on the identified content comprises:
aggregating the identified content based on logical relationships between each of the plurality of tasks; and
generating the answer in natural language based on the aggregated content using natural language processing algorithms.

9. A system comprising:
one or more memories collectively storing computer-executable instructions; and
one or more processors, wherein the one or more processors are configured to, individually or collectively, perform an operation comprising:
receiving a query input from a user;
decomposing the query input into a plurality of tasks;
querying a knowledge graph to identify one or more relevant entities based on at least one of the plurality of tasks, wherein each of the one or more entities comprises text information related to a respective portion of a document;
searching a vector database to identify one or more text chunks, wherein each of the one or more text chunks comprises contextual information in natural language for a respective identified relevant entity and is linked to the respective identified relevant entity in the knowledge graph using a chunk identifier;
identifying content relevant to at least one of the plurality of tasks by parsing the contextual information in natural language within the one or more text chunks; and
generating an answer for the query based on the identified content.

10. The system of claim 9 , wherein each of the one or more text chunks comprises text extracted from a corpus of unstructured documents.

11. The system of claim 9 , wherein each respective text chunk is vectorized into a respective vector representation using an embedding model.

12. The system of claim 9 , wherein the content relevant to the plurality of tasks from the one or more text chunks is identified using a vector-based similarity technique.

13. The system of claim 12 , wherein the vector-based similarity technique used to identify the content comprises comparing at least one of (i) cosine similarity or (ii) distance similarity between vector representations of the one or more text chunks and each of the plurality of tasks.

14. The system of claim 9 , wherein the computer-executable instructions, when executed by the one or more processors, further cause the system to:
receive feedback from the user regarding an accuracy of the answer; and
update the knowledge graph based on the feedback.

15. The system of claim 14 , wherein the computer-executable instructions, when executed by the one or more processors, further cause the system to add one or more new entities to the knowledge graph based on the feedback, wherein each of the one or more new entities comprises data derived from the answer.

16. The system of claim 9 , wherein, to generate the answer based on the identified content, the computer-executable instructions, when executed by the one or more processors, further cause the system to:
aggregate the identified content based on logical relationships between each of the plurality of tasks; and
generate the answer in natural language based on the aggregated content using natural language processing algorithms.

17. One or more computer-readable storage media containing, in any combination, computer program code that, when executed by operation of a computer system, performs operations comprising:
receiving a query input from a user;
decomposing the query input into a plurality of tasks;
querying a knowledge graph to identify one or more relevant entities based on at least one of the plurality of tasks, wherein each of the one or more entities comprises text information related to a respective portion of a document;
searching a vector database to identify one or more text chunks, wherein each of the one or more text chunks comprises contextual information in natural language for a respective identified relevant entity and is linked to the respective identified relevant entity in the knowledge graph using a chunk identifier;
identifying content relevant to at least one of the plurality of tasks by parsing the contextual information in natural language within the one or more text chunks; and
generating an answer for the query based on the identified content.

Assignees

• IBM

Inventors

• Chen Ming Jin
• Li Yuan
• Liu Tong
• Li yuan yuan
• Zhang Yuan

Classifications

• G06F16/9024

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06F16/3329

  Natural language query formulation · CPC title

• G06F16/3344 Primary

  using natural language analysis · CPC title