Artificial intelligence feature extraction service for products

• US 10,438,264 B1
• Filed: 08/31/2016
• Issued: 10/08/2019
• Est. Priority Date: 08/31/2016
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Viswanathan, Krishnamurthy Koduvayur
• Primary Examiner(s): Airapetian, Mila
• Application Number:
  US15/253,214
• Time in Patent Office: 1,133 Days
• Field of Search: None
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/252 between a Database Management System and a front-end application
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06Q 30/0623 Item investigation

Abstract

Embodiments disclosed herein include artificial intelligence systems and methods for automatically filling or completing required attribute values for features associated with an item that is being offered. In embodiments, an offer for an item may include an image of the item and unstructured text related to the item. Potential candidate values are selected and ranked according to generated confidence scores that represent the correct candidate value for the selected feature. A user interface may be presented to the user offering the item that presented the ranked potential candidate values and request confirmation. In response to the user confirming the candidate value for the feature a web page may be generated for offering the item with appropriate features and attribute values auto-populated thus providing an efficient method for offering items on behalf of a user.

Claims

1. A computer-implemented artificial intelligence method, comprising:

receiving, by a computer system, an image of an item and corresponding text for the item from a user device of a user;

determining, by the computer system, one or more features for the item that are required to offer the item in an electronic marketplace based at least in part on an item category associated with the item and a threshold amount of features associated with the item category;

identifying, by the computer system, a plurality of candidate values for a feature of the one or more features based at least in part on the corresponding text;

generating, by the computer system using a machine learning algorithm and a model, a confidence score for an individual candidate value of the plurality of candidate values for the feature;

transmitting, by the computer system to a user interface of the user device, the image of the item and a particular candidate value of the plurality of candidate values for the feature based in part on the corresponding confidence score, the particular candidate value associated with a portion of candidate values;

receiving, by the computer system and from the user interface of the user device, an indication that identifies confirmation, by the user, that the particular candidate value is correct for the feature;

in response to the user confirming, via the user interface, that the particular candidate value is correct for the feature;

generating, by the computer system, a web page that comprises the image, the particular candidate value, and the feature, the web page configured to offer the item in the electronic marketplace; and

updating, by the computer system, the model based in part on the received indication from the user interface of the user device.

2. The computer-implemented method of claim 1, wherein the machine learning algorithm generates the model that is used to generate the confidence score.

3. The computer-implemented method of claim 2, further comprising selecting, by the computer system, a particular model generated by the machine learning algorithm for generating the confidence score based in part on the item category.

4. The computer-implemented method of claim 2, further comprising in response to the user confirming, via the user interface, that the particular candidate value is incorrect for the feature:

selecting, by the computer system, a different model for generating the confidence score in response to receiving subsequent items from the item category.

5. The computer-implemented method of claim 1, further comprising in response to the user confirming, via the user interface, that the particular candidate value is incorrect for the feature:

requesting, by the computer system and via the user interface, input from the user, the input utilized as the candidate value for the feature.

6. A computer-implemented method, comprising:

a step for receiving, by a first computer system, an image from a second computer system;

a step for identifying, by the first computer system, an item in the image based in part on image recognition using information corresponding to a catalog of items maintained by the computer system;

a step for determining, by the first computer system, one or more features for the item based at least in part on an item category associated with the item and a threshold amount of features associated with the item category;

a step for identifying, by the first computer system, one or more candidate values for a feature of the one or more features for the item based at least in part on the information and a model generated by a machine learning algorithm; and

a step for transmitting, by the first computer system to the second computer system, the image of the item and a ranked portion of candidate values of the one or more candidate values for the feature, the ranked portion of the candidate values ranked based on confidence scores associated with each candidate value of the portion of candidate values.

7. The computer-implemented method of claim 6, further comprising a step for, in response to receiving input from the second computer system, generating a web page to offer the item identified in the image using a particular candidate value of the ranked portion of candidate values for the feature, the particular candidate value selected based in part on the input.

8. The computer-implemented method of claim 7, wherein the input received from the second computer system is verified based in part on image characteristics obtained, by the first computer system, of the item in the image.

9. The computer-implemented method of claim 6, wherein transmitting the image of the item and the ranked portion of candidate values for the feature is based in part on a type of the second computer system used to provide the image to the first computer system, a number of the ranked portion of candidate values selected based on the type of the second computer system.

10. The computer-implemented method of claim 6, wherein a user that is associated with the second computer system is further associated with a weighting factor by an electronic marketplace, the weighting factor determined based on previous responses by the user to the electronic marketplace.

11. The computer-implemented method of claim 10, wherein the model is updated accordingly based on the weighting factor of the user that provides the input.

12. The computer-implemented method of claim 6, further comprising a step for transmitting, by the first computer system and to the second computer system, confirmation that the item is being offered in an electronic marketplace.

13. A system, comprising:

a processor;

a speaker component;

a voice-activated component configured to receive audio; and

a memory device comprising computer-executable instructions that, when executed by the processor, cause the system to;

receive, via the voice-activated component, a voice command about an item;

identify the item from an item catalog based in part on the voice command;

identify one or more features for the item based in part on an item category for the item and a threshold amount of features associated with the item category;

transmit, via the speaker component, a series of queries, the series of queries comprising one or more candidate values for a feature of the one or more features, the one or more candidate values determined based in part on first information about the item;

receive, via the voice-activated component, individual responses to the series of queries; and

generate a web page for offering the item that comprises an image of the item and a particular candidate value of the one or more candidate values for the feature based at least in part on the individual responses.

14. The system of claim 13, wherein the one or more candidate values for the feature are determined based in part on second information from a manufacturer of the item.

15. The system of claim 13, wherein the one or more candidate values for the feature are determined based in part on second information maintained by an electronic marketplace, the second information comprising reviews of the item.

16. The system of claim 13, wherein the one or more candidate values for the feature are determined based in part on second information maintained by an electronic marketplace, the second information comprising responses provided by a plurality of users of the electronic marketplace in response to queries provided by a subset of the plurality of users.

17. The system of claim 13, wherein receiving the voice command about the item comprises using a speech-to-text conversion that generates a string of characters.

18. The system of claim 17, wherein identifying the item from the item catalog comprises using the string of characters and a corpus of item descriptions included in the item catalog.

19. The system of claim 13, wherein the series of queries are ranked based in part on a machine learning algorithm that generates a confidence score for each query of the series of queries, the confidence score representing a match for the candidate value for the feature.