Classification of query text to generate relevant query results

Patent metadata

Field	Value
Publication number	US-12293403-B2
Application number	US-202117337183-A
Country	US
Kind code	B2
Filing date	Jun 2, 2021
Priority date	Mar 27, 2019
Publication date	May 6, 2025
Grant date	May 6, 2025

Abstract

A service may provide an e-commerce platform through which users may search for, view, and purchase products. To ensure users receive relevant results when querying for particular products or product classes, a list of ranked classes may be generated, from which the query results are based. For example, products may be classified into classes and particular text from search data may be associated with one or more of the products based on click log data. For instance, input text from previously performed queries may be associated with products interacted with following those queries (e.g., products clicked on to view/purchase). Training data that associates the particular text with the respective one or more classes into which the clicked on products are classified may be generated, and in response to receiving a query including the particular text as input, the list of ranked classes may be generated based on the training data.

First claim

The invention claimed is:

1. A method of classifying text, the method comprising:
identifying a plurality of classes;
receiving user interaction data associated with a particular text;
associating the particular text with at least one of the plurality of classes based on the user interaction data;
generating text-to-class data based on the association;
providing the text-to-class data as training data to an N-gram convolutional neural network; and
in response to receiving new text for classification, providing the new text as input to the N-gram convolutional neural network to generate a subset of classes from the plurality of classes for classifying the new text, the N-gram convolutional neural network building a plurality of N-grams comprised of multiple N-gram levels to analyze the new text using the multiple N-gram levels in conjunction with the training data to output the subset of classes.

2. The method of claim 1 , wherein the subset of classes includes more than one of the plurality of classes and the method further comprises:
ranking classes within the subset based on the training data.

3. The method of claim 2 , wherein ranking the classes within the subset comprises:
calculating a class score for each class within the subset based on the training data; and
ranking the classes within the subset based on the class score.

4. The method of claim 1 , wherein the plurality of N-grams comprised of the multiple N-gram levels built by the N-gram convolutional neural network include N-grams of at least two different N-gram sizes.

5. The method of claim 1 , wherein the new text is comprised of a plurality of words, and each N-gram built by the N-gram convolutional neural network represents a different combination of one or more words from the plurality of words.

6. The method of claim 1 , wherein the new text received for classification and provided as input to the N-gram convolutional neural network is search query text, and the method further comprises:
generating search query results based on the subset of classes generated and output by the N-gram convolutional neural network.

7. A system for classifying text, the system comprising:
an N-gram convolutional neural network;
a processor; and
a memory communicatively connected to the processor, the memory storing instructions that, when executed by the processor, cause the system to:
identify a plurality of classes;
receive user interaction data associated with a particular text;
associate the particular text with at least one of the plurality of classes based on the user interaction data;
generate text-to-class data based on the association;
provide the text-to-class data as training data to the N-gram convolutional neural network; and
in response to receiving new text for classification:
provide the new text as input to the N-gram convolutional neural network to generate a subset of classes from the plurality of classes for classifying the text, the N-gram convolutional neural network building a plurality of N-grams comprised of multiple N-gram levels to analyze the new text using the multiple N-gram levels in conjunction with the training data to output the subset of classes; and
rank classes within the subset based on the training data.

8. The system of claim 7 , wherein:
the user interaction data associated with the particular text includes search data comprising a search query that includes the particular text and click log data associated with one or more of a plurality of products clicked on following the search query;
each of the one or more of the plurality of products are classified into one or more of the plurality of classes; and
to associate the particular text with at least one of the plurality of classes based on the user interaction data, the system is caused to associate the particular text with the one or more of the plurality of classes into which each of the one or more of the plurality of products are classified.

9. The system of claim 8 , wherein to generate the text-to-class data based on the association, the system is caused to one of:
associate the particular text with a single class from the one or more of the plurality of classes into which each of the one or more of the plurality of products are classified, the single class having a highest class score from among the one or more of the plurality of classes;
associate the particular text with each of the one or more of the plurality of classes into which each of the one or more of the plurality of products are classified, wherein the association is unweighted; and
associate the particular text with each of the one or more of the plurality of classes into which each of the one or more of the plurality of products are classified, wherein the association is weighted based on a class score.

10. The system of claim 8 , wherein the new text received for classification and provided as input to the N-gram convolutional neural network is search query text associated with the plurality of products, and the system is further caused to generate and provide for display search query results based on the ranked classes within the subset.

11. The system of claim 10 , wherein one or more of the plurality of products responsive to the search query text and classified into a highest ranked class within the subset are displayed first in the search query results.

12. The system of claim 7 , wherein the plurality of N-grams comprised of the multiple N-gram levels built by the N-gram convolutional neural network include N-grams of at least two different N-gram sizes.

13. The system of claim 7 , wherein the new text is comprised of a plurality of words and each N-gram built by the N-gram convolutional neural network represents a different word or combination of words from the plurality of words.

14. A method of classifying text, the method comprising:
receiving, at a convolutional neural network, training data generated based on associations of particular text with one or more of a plurality of classes;
receiving new text for classification as input to the convolutional neural network;
building a plurality of N-grams comprised of multiple N-gram levels based on the new text;
analyzing the new text using the multiple N-gram levels in conjunction with the training data to generate a subset of classes from the plurality of classes for classifying the new text; and
providing the subset of classes as output of the convolutional neural network.

15. The method of claim 14 , wherein the training data received includes text-to-class data, and the text-to-class data is generated by:
identifying the plurality of classes;
receiving user interaction data associated with the particular text;
associating the particular text with at least one of the plurality of classes based on the user interaction data; and
generating the text-to-class data based on the association.

16. The method of claim 14 , wherein the plurality of N-grams comprised of the multiple N-gram levels include N-grams of at least two different N-gram sizes.

17. The method of claim 16 , wherein an N-gram size for an N-gram of the plurality of N-grams includes at least one of: a unigram, a bigram, a trigram, a quadgram, and a pentagram.

18. The method of claim 14 , wherein the new text is comprised of a plurality of words, each N-gram represents a different combination of one or more words from the plurality of words, and each N-gram of a different N-gram size represents a different number of the one or more words.

19. The method of

Assignees

• Target Brands Inc

Inventors

• Srinivasan Vikram
• Seshadri Srinivasan
• Wang Richard

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06Q30/0625

  by formulating product or service queries, e.g. using keywords or predefined options · CPC title

• G06F40/30

  Semantic analysis · CPC title