Enhanced search result generation using multi-document summarization

Patent metadata

Field	Value
Publication number	US-12561375-B2
Application number	US-202418443838-A
Country	US
Kind code	B2
Filing date	Feb 16, 2024
Priority date	Feb 17, 2023
Publication date	Feb 24, 2026
Grant date	Feb 24, 2026

Abstract

Enhanced search results are generated using multi-document summarization. A multi-document summarization system receives a search query from a user and retrieves a plurality of search result documents based on the search query. The summarization system generates a summary of each of the plurality of search result documents using distinct per-document summarization machine learning models, where the distinct per-document summarization machine learning models are trained on a training dataset. The summarization system synthesizes the summary of each of the plurality of search result documents into a single-consolidated answer responsive to the received search query. The multi-document summarization system formats the single-consolidated answer to include citations to the plurality of search result documents.

First claim

What is claimed is:

1 . A system comprising:
one or more hardware processors of a machine; and at least one memory storing instructions that, when executed by the one or more hardware processors, cause the system to perform operations comprising:
receiving a search query;
retrieving, by at least one hardware processor, a plurality of search result documents based on the search query;
generating a summary of each of the plurality of search result documents using distinct per-document summarization machine learning models, the distinct per-document summarization machine learning models being fine-tuned using one or more summarization-specific datasets to perform both extractive summarization and using labels generated by comprehensive models that are larger than the distinct per-document summarization machine learning models;
synthesizing the generated summary of each of the plurality of search result documents into a single-consolidated answer responsive to the received search query, the synthesizing comprising performing cross-document attributed summarization of the generated summary of each of the plurality of search result documents while maintaining explicit linkages to source materials and while recognizing and excluding duplicative content from the generated summary of each of the plurality of search result documents, the performing cross-document attributed summarization of the generated summary of each of the plurality of search result documents comprising:
selecting a representative sentence from the generated summary of each of the plurality of search result documents for each common particular content identified across the plurality of search result documents; and
combining multiple selected representative sentences for each of the common particular content into the single-consolidated answer response using a unified generative artificial intelligence component that synthesizes information from disparate sources while preserving original context;

formatting the single-consolidated answer to include citations to the plurality of search result documents; and
presenting, in an interactive interface, the single-consolidated answer without requiring navigation to the plurality of search result documents.

2 . The system of claim 1 , the operations comprising:
identifying a set of documents from the plurality of search result documents pertinent to a subject of the search query; employing an additional machine learning model to discern particular content within the identified set of documents; and identifying common particular content across the plurality of search result documents.

3 . The system of claim 1 , wherein summarization of multiple documents is performed using a hierarchical approach, first summarizing individual sections within each document and second summarizing collective sections to form the single-consolidated answer.

4 . The system of claim 1 , the operations comprising:
identifying, in an automatic manner, a citation for source documents for extracted information; and embedding a hyperlink to the citation within the single-consolidated answer at a location corresponding to the extracted information.

5 . The system of claim 1 , the operations comprising:
performing real-time updates to the search query based on detection of new information relevant to the search query, wherein the real-time updates are incorporated into the single-consolidated answer without user intervention.

6 . The system of claim 1 , the operations comprising:
supporting multi-turn disambiguation by presenting a set of follow-up questions to a user based on the single-consolidated answer; receiving user responses to the set of follow-up questions; and refining the single-consolidated answer based on the user responses to the set of follow-up questions.

7 . The system of claim 1 , the operations comprising:
presenting, via a web browser on a user device, the single-consolidated answer; and receiving user input for query refinement, wherein the single-consolidated answer is dynamically updated based on the user input for the query refinement.

8 . The system of claim 1 , the operations comprising:
using one or more asymmetric compression techniques that reduce computational resources used by the distinct per-document summarization machine learning models and that enable the distinct per-document summarization machine learning models to handle concurrent user queries.

9 . A method comprising:
receiving a search query; retrieving, by at least one hardware processor, a plurality of search result documents based on the search query; generating a summary of each of the plurality of search result documents using distinct per-document summarization machine learning models, the distinct per-document summarization machine learning models being fine-tuned using one or more summarization-specific datasets to perform both extractive summarization and using labels generated by comprehensive models that are larger than the distinct per-document summarization machine learning models; synthesizing the generated summary of each of the plurality of search result documents into a single-consolidated answer responsive to the received search query, the synthesizing comprising performing cross-document attributed summarization of the generated summary of each of the plurality of search result documents while maintaining explicit linkages to source materials and while recognizing and excluding duplicative content from the generated summary of each of the plurality of search result documents, the performing cross-document attributed summarization of the generated summary of each of the plurality of search result documents comprising:
selecting a representative sentence from the generated summary of each of the plurality of search result documents for each common particular content identified across the plurality of search result documents; and
combining multiple selected representative sentences for each of the common particular content into the single-consolidated answer response using a unified generative artificial intelligence component that synthesizes information from disparate sources while preserving original context;
formatting the single-consolidated answer to include citations to the plurality of search result documents; and presenting, in an interactive interface, the single-consolidated answer without requiring navigation to the plurality of search result documents.

10 . The method of claim 9 , further comprising:
identifying a set of documents from the plurality of search result documents pertinent to a subject of the search query; employing an additional machine learning model to discern particular content within the identified set of documents; and identifying common particular content across the plurality of search result documents.

11 . The method of claim 9 , wherein summarization of multiple documents is performed using a hierarchical approach, first summarizing individual sections within each document and second summarizing collective sections to form the single-consolidated answer.

12 . The method of claim 9 , further comprising:
identifying, in an automatic manner, a citation for source documents for extracted information; and embedding a hyperlink to the citation within the single-consolidated answer at a location corresponding to the extracted information.

13 . The method of claim 9 , further comprising:
performing real-time updates to the search query based on detection of new information relevant to the search query, wherein the real-time updates are incorporated into the single-consolid

Assignees

• Snowflake Inc

Inventors

• Bathwal Rahil
• Campos Daniel Fernando
• Devaraj Ashwin
• Li Seth Michael
• Ngan Muhua
• Raghunathan Vivek
• Ramaswamy Sridhar
• Samdani Rajhans
• So Chiu Wah
• Tarakad Nitya Kannan

Classifications

• G06F16/248

  Presentation of query results · CPC title

• G06F16/24575 Primary

  using context · CPC title

• G06F16/9538

  Presentation of query results · CPC title

• G06F16/9558

  Details of hyperlinks; Management of linked annotations · CPC title

• G06F16/93

  Document management systems · CPC title