IP Library Granted Patent US 12,469,068
Granted Patent B2
US 12,469,068 · App. 18/545,452 · Granted Nov 11, 2025

Search results summarization tuning

Inventors: Soumya Unnikrishnan (Austin, TX); Saina Lajevardi (Austin, TX); Michele Saad (Austin, TX)
Assignee: ADOBE INC.
G06Q30/0631G06F40/247G06F40/40G06Q30/0204G06Q30/0629
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,469,068
App. No.
18/545,452
Granted
Nov 11, 2025
Kind
B2
Abstract

One or more aspects of the method, apparatus, and non-transitory computer readable medium include receiving a query relating to an item and a summarization type indicating an emphasis on item similarities or item differences, obtaining, using a search component, descriptions of items relevant to the query, generating input data for a machine learning model based on the descriptions and the summarization type, and generating, using the machine learning model, a summarization of the descriptions based on the input data in response to the query, wherein the summarization emphasizes the item similarities or item differences based on the summarization type.

Claims (54)

1 . A method comprising:

receiving a query relating to an item and a summarization type indicating an emphasis on item similarities or item differences;

obtaining, using a search component, descriptions of items relevant to the query;

generating input data for a machine learning model based on the descriptions and the summarization type; and

generating, using the machine learning model, a summarization of the descriptions based on the input data in response to the query, wherein the summarization emphasizes the item similarities or the item differences based on the summarization type.

2 . The method of claim 1 , further comprising:

tuning the summarization to include topics learned from a user.

3 . The method of claim 1 , further comprising:

processing the query using the machine learning model to identify search terms; and

conducting a search based on the query to identify items relevant to the search terms, wherein descriptions associated with the items are used by the machine learning model to generate questions for a user.

4 . The method of claim 1 , wherein:

the generated summarization is a comparative summarization that suppresses repeated information in the descriptions and accentuates similarities and differences between the items.

5 . The method of claim 1 , wherein:

generating follow-up questions for a user using the machine learning model to identify new search topics.

6 . The method of claim 5 , further comprising:

conducting a subsequent search based on the new search topics.

7 . The method of claim 1 , wherein:

a conciseness of the summarization of the descriptions is based on a compression parameter that determines.

8 . A method, comprising:

receiving a ranked list of items from a search component as a result of an initial search query;

presenting a plurality of ideation questions generated by a machine learning model to a user, wherein the plurality of ideation questions are generated based on the ranked list of items and the initial search query; and

conducting, using the search component, a subsequent search using a new query generated by the machine learning model based on responses provided by the user to the generated ideation questions.

9 . The method of claim 8 , wherein:

the user selects one or more ideation questions from the plurality of ideation questions to further refine a search space.

10 . The method of claim 8 , wherein:

the plurality of ideation questions are generated based on topics identified by the user through the responses provided by the user.

11 . The method of claim 8 , further comprising:

optimizing a distribution of the plurality of generated ideation questions using a multiarmed bandit approach.

12 . The method of claim 8 , further comprising:

identifying topics from the responses provided by the user to the plurality of generated ideation questions; and

incorporating the identified topics into the new query.

13 . The method of claim 12 , further comprising:

conducting the subsequent search using the new query generated based on the identified topics.

14 . The method of claim 8 , further comprising:

aggregating descriptive content for the ranked list of items; and

providing the aggregated descriptive content to a summarization component as an input to generate the ideation questions.

15 . A system comprising:

one or more processors;

one or more memories including instructions executable by the one or more processors to:

receive an initial search query relating to an item and a summarization type indicating an emphasis on item similarities or item differences;

obtain, using a search component, a ranked list of items, as a result of the initial search query;

generate, using a machine learning model, a comparative summarization of descriptions for the ranked list of items based on the summarization type; and

present the comparative summarization and a plurality of generated ideation questions to a user, wherein the plurality of ideation questions are generated by an ideation component based on the ranked list of items and the initial search query.

16 . The system of claim 15 , wherein:

the machine learning model includes a transformer encoder and a transformer decoder.

17 . The system of claim 15 , wherein:

the machine learning model is a pre-trained large language model (LLM) configured to generate the comparative summarization of descriptions based on a vector embedding.

18 . The system of claim 15 , further comprising:

receiving, by the machine learning model, responses to the plurality of ideation questions from the user; and

conducting a subsequent search using a new query generated by the machine learning model based on the responses.

19 . The system of claim 18 , wherein:

the ideation component utilizes a multi-armed bandit approach to identify the plurality of generated ideation questions for presentation to the user from a larger set of generated ideation questions.

20 . The system of claim 19 , wherein:

the machine learning model is trained based on supervised contrastive learning.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 19, 2023
From: UNNIKRISHNAN, SOUMYA; LAJEVARDI, SAINA; SAAD, MICHELE
To: ADOBE, INC.
Reel/Frame 065913/0029 →
Continuity (1)
Related Publication 20250200635A1 · Jun 19, 2025
References Cited (8)
US 10296540B1 · Bhole · 2019 [cited by examiner]
US 20200019419A1 · Yada · 2020 [cited by examiner]
US 20200142978A1 · Salokhe · 2020 [cited by examiner]
US 20200349214A1 · Tang · 2020 [cited by examiner]
US 20210110464A1 · Tufegdzic · 2021 [cited by examiner]
Brown, et al, “Language Models are Few-Shot Learners”, arXiv preprint: arXiv:2005.14165v4 [cs.CL] Jul. 22, 2020, 75 pages. [cited by applicant]
Manning, “Human Language Understanding & Reasoning”, Daedalus (2022) 151 (2): 127-138, May 1, 2022, 12 pages. [cited by applicant]
Nan, et al, “Improving Factual Consistency of Abstractive Summarization via Question Answering”, arXiv preprint: arXiv:2105.04623v1 [cs.CL] May 10, 2021, 16 pages. [cited by applicant]