Responsive category prediction for user queries

Patent metadata

Field	Value
Publication number	US-2022100806-A1
Application number	US-202117487732-A
Country	US
Kind code	A1
Filing date	Sep 28, 2021
Priority date	Sep 30, 2020
Publication date	Mar 31, 2022
Grant date	—

Abstract

A method for determining a category responsive to a user query is disclosed. The method includes receiving a training data set comprising a plurality of data pairs, each data pair including: (i) a query; and (ii) an associated one or more categories that are responsive to the query, wherein the one or more categories in the training data set defines a plurality of categories. The method includes training a machine learning algorithm, according to the training data set, to create a trained model, wherein training the machine learning algorithm includes: creating a first co-occurrence data structure defining co-occurrence of respective word representations of the queries with the plurality of categories, and creating a second co-occurrence data structure defining co-occurrence of respective categories in respective data pairs. The method also includes deploying the trained model to return one or more categories in response to a new query input.

First claim

What is claimed is:

1 . A method for determining a category responsive to a user query, the method comprising:
receiving a training data set comprising a plurality of data pairs, each data pair comprising: (i) a query; and (ii) an associated one or more categories that are responsive to the query, wherein the one or more categories in the training data set defines a plurality of categories; training a machine learning algorithm, according to the training data set, to create a trained model, wherein training the machine learning algorithm comprises:
creating a first co-occurrence data structure defining co-occurrence of respective word representations of the queries with the plurality of categories; and
creating a second co-occurrence data structure defining co-occurrence of respective categories in respective data pairs; and
deploying the trained model to return one or more categories in response to a new query input.

2 . The method of claim 1 , wherein training the machine learning algorithm further comprises:
separating each query into a respective one or more words that comprise the query; and calculating respective embeddings for each of the one or more words to create a word embeddings set; wherein the word embeddings set comprises the word representations.

3 . The method of claim 2 , wherein training the machine learning algorithm further comprises:
inputting the first co-occurrence data structure to a self-attention mechanism, wherein the self-attention mechanism outputs a relative correlation between each category and each word representation; and applying the relative correlation between each category and each word representation as a weight set to the word embeddings set.

4 . The method of claim 1 , wherein training the machine learning algorithm further comprises:
calculating respective embeddings for each of the plurality of categories to create a category embeddings set; wherein defining co-occurrence of respective categories in respective data pairs comprises defining co-occurrence of embeddings respective of categories in respective data pairs.

5 . The method of claim 1 , wherein training the machine learning algorithm further comprises one or more of:
minimizing a loss in the first co-occurrence data structure in successive training iterations; or minimizing a loss in the second co-occurrence data structure in successive training iterations.

6 . The method of claim 1 , wherein training the machine learning algorithm further comprises minimizing a combined loss in the first co-occurrence data structure and in the second co-occurrence data structure in successive training iterations.

7 . The method of claim 1 , wherein deploying the trained model to return one or more categories in response to a new query input comprises:
receiving the new user query through an electronic interface; inputting the new user query to the trained model; and outputting the output of the trained model to the user through the electronic interface.

8 . A method for determining a category responsive to a user query, the method comprising:
receiving a training data set comprising a plurality of data pairs, each data pair comprising: (i) a query; and (ii) an associated one or more categories that are responsive to the query; training a machine learning algorithm, according to the training data set, to create a trained model, wherein training the machine learning algorithm comprises:
defining a first predictive relationship of respective queries to respective categories;
defining a second predictive relationship of respective categories to one another; and
minimizing a combined loss of the first predictive relationship and the second predictive relationship; and
deploying the trained model to return one or more categories in response to a new query input.

9 . The method of claim 8 , wherein defining the first predictive relationship comprises creating a first co-occurrence data structure, based on the training data, that defines co-occurrence of respective word representations of the queries with the plurality of categories.

10 . The method of claim 8 , wherein defining the second predictive relationship comprises creating a second co-occurrence data structure, based on the training data, that includes co-occurrence of respective categories in respective data pairs.

11 . The method of claim 8 , wherein training the machine learning algorithm further comprises:
separating each query into a respective one or more words that comprise the query; calculating respective embeddings for each of the one or more words to create a word embeddings set; determining a relative correlation between each category and respective word representations of each query in the training data; and applying the relative correlation between each category and each word representation as a weight set to the word embeddings set.

12 . The method of claim 8 , wherein defining the first predictive relationship of respective queries to respective categories comprises:
separating each query in the training data set into a respective one or more words that comprise the query; and calculating respective embeddings for each of the one or more words to create a word embeddings set; wherein the word embeddings set comprises the word representations.

13 . The method of claim 8 , further comprising:
calculating respective embeddings for each of the categories to create a category embeddings set; wherein defining the second predictive relationship of respective categories to one another comprises comprises defining a predictive relationship of embeddings respective of categories in data pairs.

14 . The method of claim 8 , wherein deploying the trained model to return one or more categories in response to a new query input comprises:
receiving the new user query through an electronic interface; inputting the new user query to the trained model; and outputting the output of the trained model to the user through the electronic interface.

15 . A method for determining a category responsive to a user query, the method comprising:
receiving the user query through an electronic interface; inputting the user query to a trained machine learning model, wherein the trained model:
decomposes the new user query into one or more words;
calculates word embeddings of the one or more words;
calculates, in response to receiving the user query, category embeddings of one or more potentially-responsive categories; and
determines one or more of the potentially-responsive categories that are most likely to be responsive to the user query according to the word embeddings and the category embeddings; and
outputting the one or more potentially-responsive categories that are most likely to be responsive to the user query to the user through the electronic interface.

16 . The method of claim 15 , further comprising:
inputting the user query into a ranking algorithm to identify one or more items that are potentially responsive to the new user query, each of the one or more items associated with one or more categories; wherein calculating category embeddings of one or more potentially-responsive categories comprises calculating embeddings of the one or more categories associated with the one or more items.

17 . The method of claim 15 , wherein outputting the one or more potentially-responsive categories that are most likely to be responsive to the user query to the user through the electronic interface comprises outputting one or more items that are within th

Assignees

• Home Depot Product Authority Llc

Inventors

• Ahmadvand Ali
• Kallumadi Surya
• Javed Faizan

Classifications

• G06F18/2148

  characterised by the process organisation or structure, e.g. boosting cascade · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title