Multi-modal machine-learning model training for search

Patent metadata

Field	Value
Publication number	US-11914665-B2
Application number	US-202217675290-A
Country	US
Kind code	B2
Filing date	Feb 18, 2022
Priority date	Feb 18, 2022
Publication date	Feb 27, 2024
Grant date	Feb 27, 2024

Abstract

Multi-modal machine-learning model training techniques for search are described that overcome conventional challenges and inefficiencies to support real time output, which is not possible in conventional training techniques. In one example, a search system is configured to support multi-modal machine-learning model training. This includes use of a preview mode and an expanded mode. In the preview mode, a preview segment is generated as part of real time training of a machine learning model. In the expanded mode, the preview segment is persisted as an expanded segment that is used to train and utilize an expanded machine-learning model as part of search.

First claim

What is claimed is:

1. In a digital medium machine-learning model training environment for search, a method implemented by a computing device, the method comprising:
receiving, by the computing device, an input requesting generation of a preview segment from a base segment;
displaying, by the computing device in real time responsive to the input, a base accuracy/reach graph in a user interface, the displaying including:
generating base training data by sampling corresponding usage data from a base set of entities, defined by the base segment, taken from a plurality of entities included in a data lake;
training a base machine-learning model using the base training data;
generating search results, using the base machine-learning model, indicating event occurrence probabilities for the plurality of entities; and
generating the base accuracy/reach graph from the search results;

generating, by the computing device, the preview segment based on a user input specifying an amount of reach or accuracy via the base accuracy/reach graph; and
controlling, by the computing device, digital service operation of as part of an executable service platform of a service provider system based on a preview machine-learning model trained for the preview segment.

2. The method as described in claim 1 , wherein the entities are devices and the controlling the digital service operation includes controlling operation of the devices based on inclusion in the preview segment and respective probabilities of having the event occurrence as determined by the preview machine-learning model.

3. The method as described in claim 1 , wherein the event occurrence involves device operation as part of the executable service platform of the service provider system.

4. The method as described in claim 1 , wherein the controlling the digital service operation includes controlling access to digital content for the plurality of entities.

5. The method as described in claim 1 , wherein the plurality of entities are specified as accounts implemented by the service provider system to access the digital services via a network.

6. The method as described in claim 1 , wherein the displaying the base accuracy/reach graph in the user interface by the computing device in real time responsive to the input is performed within ten seconds of receiving the input.

7. The method as described in claim 1 , wherein the preview segment includes at least one additional said entity that is not included in the base segment.

8. The method as described in claim 1 , further comprising:
generating, by the computing device, preview training data by sampling corresponding usage data from a preview set of entities, defined by the preview segment, taken from the plurality of entities included in the data lake;
training, by the computing device, the preview machine-learning model using the base training data; and
generating, by the computing device, search results, using the preview machine-learning model, indicating event occurrence probabilities for the preview set of entities.

9. The method as described in claim 1 , further comprising:
receiving, by the computing device, a user input to persist the preview segment as an expanded segment;
receiving, by the computing device, an input scheduled to initiate training of an expanded machine-learning model;
generating, by the computing device, the expanded segment based on a segment definition from the preview segment persisted in storage responsive to the user input;
generating, by the computing device, expanded training data for an expanded set of the plurality of entities corresponding to the expanded segment;
training, by the computing device, an expanded machine-learning model utilizing the preview training data as part of machine learning to predict the event occurrence; and
generating, by the computing device, search results using the expanded machine-learning model.

10. In a digital medium training environment, a system comprising:
a multi-mode selection module implemented by a processor to output a preview mode representation and an expanded mode representation;
a preview mode training module implemented by the processor to generate a preview segment responsive to selection of the preview mode representation, the preview mode training module including:
a sampling module to generate base training data by sampling corresponding usage data from a base set of entities, defined by a base segment, taken from a plurality of entities included in a data lake;
a search module to generate search results using a base machine-learning model trained using the base training data, the search results indicating event occurrence probabilities for the plurality of entities;
a user interface module to receive, via a user interface, a user input specifying an amount of reach or accuracy through interaction with a base accuracy/reach graph generated from the search results in real time; and
a segment generation module to generate the preview segment based on the user input and the base segment; and

an expanded mode training module configured to persist the preview segment as an expanded segment through scheduled training of an expanded machine-learning model responsive to selection of the expanded mode representation.

11. The system as described in claim 10 , further comprising a model training module to train the base machine-learning model using the base training data.

12. The system as described in claim 10 , further comprising a graph generation module to generate the base accuracy/reach graph from the search results.

13. The system as described in claim 10 , wherein the expanded mode training module is configured to employ model results corresponding to a preview machine-learning model trained based on the preview segment and a segment definition corresponding to the preview segment to train the expanded machine-learning model.

14. The system as described in claim 10 , wherein the entities are devices and further comprising a resource provisioning module configured to control operation of the devices.

15. The system as described in claim 10 , further comprising a digital content access control module configured to control access to digital content based on respective probabilities of having the event occurrence as determined by a preview machine-learning model trained for the preview segment.

16. The system as described in claim 10 , wherein the preview mode training module is configured to train the base machine-learning model in less than a threshold amount of time and the expanded mode training module is configured to train an expanded machine-learning model corresponding to the expanded segment in greater than the threshold amount of time.

17. The system as described in claim 16 , wherein the threshold amount of time is ten seconds.

18. In a digital medium environment for search, a system comprising:
means for receiving an input requesting generation of a preview segment from a base segment;
means for displaying, in real time responsive to the input, a base accuracy/reach graph in a user interface, the displaying means including:
means for generating base training data by sampling corresponding usage data from a base set of entities, defined by the base segment, taken from a plurality of entities included in a data lake;
means for training a base machine-learning model using the base training data;
means for generating search results, using the base machine-learning model, indicating event occurrence probabilities for the plurality of entities; and
means fo

Assignees

• Adobe Inc

Inventors

• Kapilevich Matvey
• Savova Margarita R
• Rao Anup Bandigadi
• Mai Tung Thanh
• Shivalingaiah Lakshmi
• Snai Liron Goren
• Menguy Charles
• Lomada Vijeth
• Sinha Moumita
• Sahni Harleen

Classifications

• G06F16/9538 Primary

  Presentation of query results · CPC title

• G06F16/248

  Presentation of query results · CPC title

• G06F16/283

  Multi-dimensional databases or data warehouses, e.g. MOLAP or ROLAP · CPC title

• G06F16/9024

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

• G06N20/00 Primary

  Machine learning · CPC title