Granted Patent B1

US 12,572,374 · App. 17/892,733 · Granted Mar 10, 2026

Inventors: Mayank Tiwari (Serilingampalle, IN); Pankaj Rathoure (Hyderabad, IN); Anirban Bhattacharyya (Ammenpur, IN); Santosh Kumar (Jharkhand, IN)

Assignee: Blue Yonder Group, Inc.

G06F9/453G06F16/958G06F40/20

Abstract

Claims (45)

1 . A system for predicting recommendations for a user interface, comprising:

a computer, comprising a processor and memory, and configured to:

generate a graphical user interface that receives a conversation-based input from at least one user of a client portal;

receive a user profile and a navigation history for the at least one user;

provide the conversation-based input from the at least one user to a natural language processing engine;

decode, by the natural language processing engine, an intent and one or more slots from the conversation-based input, wherein the intent is determined according to meta-classes to be low-complexity or high-complexity, wherein the meta-classes comprise one or more of: recognize, enter, initiate, overview and select;

generate a plurality of ranked recommendations based, at least in part, on the intent, the one or more slots, the user profile, and the navigation history, wherein the plurality of ranked recommendations are generated using a probabilistic matrix factorization, wherein plurality of ranked recommendations comprise navigational flows between levels of a hierarchy of nodes and arcs which correspond to organization of the graphical user interface, wherein each node of the hierarchy of nodes corresponds to data locations, functions, actions, layouts, screens, and modules which a user may view to perform related tasks, and further wherein the plurality of ranked recommendations comprise tasks of a functional workflow;

monitor actual actions and navigation initiated by the at least one user to determine correct and incorrect recommendations; and

based on the monitoring, update a learning model to improve future recommendations and their rankings.

2 . The system of claim 1 , wherein the one or more recommendations comprise a subsequent action for the at least one user to complete at least one task, a quantity of remaining steps to complete the at least one task, and an identification of any slots and entities needed to complete the quantity of remaining steps.

3 . The system of claim 1 , wherein the one or more recommendations are ranked according to a confidence score.

4 . The system of claim 1 , wherein the decoded intent is a categorical assignment that describes a purpose or goal of the conversation-based input.

5 . The system of claim 1 , wherein the computer is configured to generate the one or more recommendations by:

crawling tasks associated with the at least one user;

indexing the crawled tasks by storing and organizing task definitions according to one or more intents that initiate a task; and

ranking the crawled tasks to identify tasks most relevant to the decoded intent.

6 . A computer-implemented method for predicting recommendations for a user interface, comprising:

generating, by a computer comprising a processor and memory, a graphical user interface that receives a conversation-based input from at least one user of a client portal;

receiving, by the computer, a user profile and a navigation history for the at least one user, providing, by the computer, the conversation-based input from the at least one user to a natural language processing engine;

decoding, by the computer using the natural language processing engine, an intent and one or more slots from the conversation-based input, wherein the intent is determined according to meta-classes to be low-complexity or high-complexity, wherein the meta-classes comprise one or more of: recognize, enter, initiate, overview and select;

generating, by the computer, a plurality of ranked recommendations based, at least in part, on the intent, the one or more slots, the user profile, and the navigation history, wherein the a plurality of ranked recommendations are generated using a probabilistic matrix factorization, wherein plurality of ranked recommendations comprise navigational flows between levels of a hierarchy of nodes and arcs which correspond to organization of the graphical user interface, wherein each node of the hierarchy of nodes corresponds to data locations, functions, actions, layouts, screens, and modules which a user may view to perform related tasks, and further wherein plurality of ranked recommendations comprise tasks of a functional workflow;

monitoring, by the computer, actual actions and navigation initiated by the at least one user to determine correct and incorrect recommendations; and

based on the monitoring, updating, by the computer, a learning model to improve future recommendations and their rankings.

7 . The computer-implemented method of claim 6 , wherein the one or more recommendations comprise a subsequent action for the at least one user to complete at least one task, a quantity of remaining steps to complete the at least one task, and an identification of any slots and entities needed to complete the quantity of remaining steps.

8 . The computer-implemented method of claim 6 , wherein the one or more recommendations are ranked according to a confidence score.

9 . The computer-implemented method of claim 6 , wherein the decoded intent is a categorical assignment that describes a purpose or goal of the conversation-based input.

10 . The computer-implemented method of claim 6 , further comprising generating the one or more recommendations by:

crawling, by the computer, tasks associated with the at least one user;

indexing, by the computer, the crawled tasks by storing and organizing task definitions according to one or more intents that initiate a task; and

ranking, by the computer, the crawled tasks to identify tasks most relevant to the decoded intent.

11 . A non-transitory computer-readable storage medium embodied with software for predicting recommendations for a user interface, the software when executed by a computer is configured to:

generate a graphical user interface that receives a conversation-based input from at least one user of a client portal;

receive a user profile and a navigation history for the at least one user;

provide the conversation-based input from the at least one user to a natural language processing engine;

decode, by the natural language processing engine, an intent and one or more slots from the conversation-based input, wherein the intent is determined according to meta-classes to be low-complexity or high-complexity, wherein the meta-classes comprise one or more of: recognize, enter, initiate, overview and select;

generate a plurality of ranked recommendations based, at least in part, on the intent, the one or more slots, the user profile, and the navigation history, wherein the plurality of ranked recommendations are generated using a probabilistic matrix factorization, wherein the plurality of ranked recommendations comprise navigational flows between levels of a hierarchy of nodes and arcs which correspond to organization of the graphical user interface, wherein each node of the hierarchy of nodes corresponds to data locations, functions, actions, layouts, screens, and modules which a user may view to perform related tasks, and further wherein the plurality of ranked recommendations comprise tasks of a functional workflow;

monitor actual actions and navigation initiated by the at least one user to determine correct and incorrect recommendations; and

based on the monitoring, update a learning model to improve future recommendations and their rankings.

12 . The non-transitory computer-readable storage medium of claim 11 , wherein the one or more recommendations comprise a subsequent action for the at least one user to complete at least one task, a quantity of remaining steps to complete the at least one task, and an identification of any slots and entities needed to complete the quantity of remaining steps.

13 . The non-transitory computer-readable storage medium of claim 11 , wherein the one or more recommendations are ranked according to a confidence score.

14 . The non-transitory computer-readable storage medium of claim 11 , wherein the decoded intent is a categorical assignment that describes a purpose or goal of the conversation-based input.

15 . The non-transitory computer-readable storage medium of claim 11 , wherein the software is configured to generate the one or more recommendations by:

crawl tasks associated with the at least one user;

index the crawled tasks by storing and organizing task definitions according to one or more intents that initiate a task; and

rank the crawled tasks to identify tasks most relevant to the decoded intent.

Assignments (1)

Continuity (2)

References Cited (47)