Granted Patent B2

US 12,541,782 · App. 17/147,939 · Granted Feb 3, 2026

Inventors: Nanyang Wang (Hangzhou, CN); Bin Wang (Hangzhou, CN); Yi Wang (Hangzhou, CN); Pan Pan (Beijing, CN); Yinghui Xu (Hangzhou, CN)

Assignee: ALIBABA GROUP HOLDING LIMITED

G06Q30/0627G06F16/483G06N3/04G06V20/20

Abstract

Claims (102)

1 . A product object publishing method, comprising:

obtaining, by a client application associated with a website, an image of a product object;

inputting, by the client application, the image to an image recognizer set to generate a plurality of types of product attribute information, wherein the image recognizer set comprises a plurality of machine learning models respectively corresponding to the plurality of types of product attribute information, and the inputting comprises:

concurrently inputting the image into the plurality of machine learning models for concurrent image recognition and outputting corresponding types of product attribute information:

training the plurality of machine learning models, the training comprising:

pulling structured information and images of published product objects from a database of the website, wherein the structured information comprises a brand, a category, a material, a condition, or a model of a published product; and

training, for each of the plurality of types of product attribute information, a machine learning model based on the pulled images and corresponding structured information for recognizing the corresponding type of product attribute information for a given image, thereby obtaining the plurality of machine learning models;

wherein one of the plurality of machine learning models comprises a plurality of sub-recognizers, and the concurrently inputting the image comprises:

concurrently inputting the image into the plurality of sub-recognizers of the one machine learning model;

determining sub-recognition results outputted by the plurality of sub-recognizers respectively, wherein the sub-recognition results comprise different portions of a type of product attribute information of the image; and

combining the sub-recognition results to obtain the type of product attribute information of the image;

receiving, by the client application, the plurality of types of product attribute information of the product object from the image recognizer set;

determining, by the client application, a plurality of attribute tags for the product object based on statistics of search keywords corresponding to the product object, wherein the determining comprises:

computing, for each of the search keywords, a ratio between a number of times a search keyword appears in association with purchases of the product object and a number of times the search keyword appears in association with views of the product object,

selecting top-N of the search keywords as the plurality of attribute tags of the product object based on the computed ratios, N being an integer greater than one, and

assigning a weight to each of the plurality of attribute tags for subsequent selection;

displaying, by the client application, the plurality of types of product attribute information and the plurality of attribute tags of the product object for selection by a user;

generating, by the client application, structured information of the product object according to selection of the product attribute information and the one or more attribute tags; and

publishing, by the client application, the product object, wherein the publishing comprises publishing the image and structured information of the product object.

2 . The method according to claim 1 , wherein the obtaining an image of a product object comprises:

displaying a publishing setting page, wherein a shooting option is provided in the publishing setting page or invoking a different application to select a pre-shot image of the product object; and

invoking, according to triggering of the shooting option, a shooting component of the client application to shoot an image of the product object.

3 . The method according to claim 1 , wherein:

the inputting the image comprises generating a recognition request according to the image of the product object, and sending the recognition request to a server; and

the receiving the plurality of types of product attribute information comprises receiving a recognition response corresponding to the recognition request, wherein the recognition response comprises the plurality of types of product attribute information of the product object.

4 . The method according to claim 3 , wherein

the recognition response further comprises association relationships among the plurality of types of product attribute information, and

the generating structured information of the product object according to selection of the product attribute information comprises:

receiving a selection instruction for target product attribute information;

determining, according to the association relationships, product attribute information associated with the target product attribute information;

displaying the product attribute information associated with the target product attribute information for selection; and

generating, based on a hierarchical structure, the structured information of the product object according to a plurality of pieces of selected product attribute information.

5 . The method according to claim 1 , further comprising:

receiving, by the client application, a custom tag from the user, and adding the custom tag to the structured information of the product object.

6 . A product object publishing system comprising:

a processor; and

a non-transitory computer-readable storage medium storing instructions executable by the processor to cause the system to perform operations comprising:

obtaining an image of a product object for publishing on a website;

inputting the image to an image recognizer set on a server to generate a plurality of types of product attribute information, wherein the image recognizer set comprises a plurality of machine learning models respectively corresponding to the plurality of types of product attribute information, and the inputting comprises:

concurrently inputting the image into the plurality of machine learning models for concurrent image recognition and outputting corresponding types of product attribute information;

training the plurality of machine learning models, the training comprising:

pulling structured information and images of published product objects from a database of the website, wherein the structured information comprises a brand, a category, a material, a condition, or a model of a published product; and

training, for each of the plurality of types of product attribute information, a machine learning model based on the pulled images and corresponding structured information for recognizing the corresponding type of product attribute information for a given image, thereby obtaining the plurality of machine learning models;

wherein one of the plurality of machine learning models comprises a plurality of sub-recognizers, and the concurrently inputting the image comprises:

concurrently inputting the image into the plurality of sub-recognizers of the one machine learning model;

determining sub-recognition results outputted by the plurality of sub-recognizers respectively, wherein the sub-recognition results comprise different portions of a type of product attribute information of the image; and

combining the sub-recognition results to obtain the type of product attribute information of the image;

receiving the plurality of types of product attribute information of the product object from the image recognizer set on the server;

determining a plurality of attribute tags for the product object based on statistics of search keywords corresponding to the product object, wherein the determining comprises:

computing, for each of the search keywords, a ratio between a number of times a search keyword appears in association with purchases of the product object and a number of times the search keyword appears in association with views of the product object,

selecting top-N of the search keywords as the plurality of attribute tags of the product object based on the computed ratios, N being an integer greater than one, and

assigning a weight to each of the plurality of attribute tags for subsequent selection;

displaying the plurality of types of product attribute information and the plurality of attribute tags of the product object for selection by a user;

generating structured information of the product object according to selection of the product attribute information and the one or more attribute tags; and

publishing the image and structured information of the product object.

7 . The product object publishing system according to claim 6 , wherein:

the inputting the image comprises generating a recognition request according to the image of the product object, and sending the recognition request to a server; and

the receiving the plurality of types of product attribute information comprises receiving a recognition response corresponding to the recognition request, wherein the recognition response comprises the plurality of types of product attribute information of the product object.

8 . The product object publishing system according to claim 7 , wherein the recognition response further comprises association relationships among the plurality of types of product attribute information, and

the generating structured information of the product object according to selection of the product attribute information comprises:

receiving a selection instruction for target product attribute information;

determining, according to the association relationships, product attribute information associated with the target product attribute information;

displaying the product attribute information associated with the target product attribute information for selection; and

generating, based on a hierarchical structure, the structured information of the product object according to a plurality of pieces of selected product attribute information.

9 . The product object publishing system according to claim 6 , wherein the obtaining an image of a product object comprises:

displaying a publishing setting page, wherein a shooting option is provided in the publishing setting page or invoking a different application to select a pre-shot image of the product object; and

invoking, according to triggering of the shooting option, a shooting component of the client application to shoot an image of the product object.

10 . The product object publishing system according to claim 6 , wherein the operations further comprise:

receiving a custom tag from the user, and adding the custom tag to the structured information of the product object.

11 . A non-transitory computer-readable storage medium configured with instructions executable by one or more processors to cause the one or more processors to perform operations comprising:

obtaining an image of a product object for publishing on a website;

inputting the image to an image recognizer set on a server to generate a plurality of types of product attribute information, wherein the image recognizer set comprises a plurality of machine learning models respectively corresponding to the plurality of types of product attribute information, and the inputting comprises:

concurrently inputting the image into the plurality of machine learning models for concurrent image recognition and outputting corresponding types of product attribute information;

training the plurality of machine learning models, the training comprising:

pulling structured information and images of published product objects from a database of the website, wherein the structured information comprises a brand, a category, a material, a condition, or a model of a published product; and

training, for each of the plurality of types of product attribute information, a machine learning model based on the pulled images and corresponding structured information for recognizing the corresponding type of product attribute information for a given image, thereby obtaining the plurality of machine learning models;

wherein one of the plurality of machine learning models comprises a plurality of sub-recognizers, and the concurrently inputting the image comprises:

concurrently inputting the image into the plurality of sub-recognizers of the one machine learning model;

determining sub-recognition results outputted by the plurality of sub-recognizers respectively, wherein the sub-recognition results comprise different portions of a type of product attribute information of the image; and

combining the sub-recognition results to obtain the type of product attribute information of the image;

receiving the plurality of types of product attribute information of the product object from the image recognizer set on the server;

determining a plurality of attribute tags for the product object based on statistics of search keywords corresponding to the product object, wherein the determining comprises:

computing, for each of the search keywords, a ratio between a number of times a search keyword appears in association with purchases of the product object and a number of times the search keyword appears in association with views of the product object,

selecting top-N of the search keywords as the plurality of attribute tags of the product object based on the computed ratios, N being an integer greater than one, and

assigning a weight to each of the plurality of attribute tags for subsequent selection;

displaying the plurality of types of product attribute information and the plurality of attribute tags of the product object for selection by a user;

generating structured information of the product object according to selection of the product attribute information and the one or more attribute tags; and

publishing the image and structured information of the product object.

12 . The non-transitory computer-readable storage medium of claim 11 , wherein:

the inputting the image comprises generating a recognition request according to the image of the product object, and sending the recognition request to a server; and

the receiving the plurality of types of product attribute information comprises receiving a recognition response corresponding to the recognition request, wherein the recognition response comprises the plurality of types of product attribute information of the product object.

13 . The non-transitory computer-readable storage medium of claim 12 , wherein the recognition response further comprises association relationships among the plurality of types of product attribute information, and

the generating structured information of the product object according to selection of the product attribute information comprises:

receiving a selection instruction for target product attribute information;

determining, according to the association relationships, product attribute information associated with the target product attribute information;

displaying the product attribute information associated with the target product attribute information for selection; and

generating, based on a hierarchical structure, the structured information of the product object according to a plurality of pieces of selected product attribute information.

14 . The non-transitory computer-readable storage medium of claim 11 , wherein the obtaining an image of a product object comprises:

displaying a publishing setting page, wherein a shooting option is provided in the publishing setting page or invoking a different application to select a pre-shot image of the product object; and

invoking, according to triggering of the shooting option, a shooting component of the client application to shoot an image of the product object.

15 . The non-transitory computer-readable storage medium of claim 11 , wherein the operations further comprise:

receiving, by the client application, a custom tag from the user, and adding the custom tag to the structured information of the product object.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (39)